Reading bench: Text
Leaderboard
| Tier | Model | Accuracyof 2,198 questions, 95% interval | Readingof 656 questions | Grammarof 1,542 questions | Non-hallucination% “I don't know” of answers not right | Confidencecalibration gap, points | Costper 1,000 questionsper 1M questions | Latencymedian s per request |
|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon high · chat‡ Closed · FlagshipLatency 217.7 sCost $31.309$31,309 per 1,000per 1M Gemini 4 Carbon high · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Accuracy 98.5% (95%: 97.8–98.9%) Reading 97.6% · grammar 98.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.9% · answered 99.7% · error rate 1.5% Confidence: calibration gap 0.4 points (written estimate) Cost $31.309$31,309 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 217.7 s median per request, 1 in flight | 98.5%97.8%–98.9% | 97.6% | 98.8% | 17.9%answered 100% | 0.4written estimate | $31.309$31,309subscription, priced as the API | 217.716 per request1 in flight |
| 1 | Claude Fable 5.2 max · chat‡ Closed · FlagshipLatency 81.1 sCost $53.381$53,381 per 1,000per 1M Claude Fable 5.2 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $8.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-11-11) Accuracy 98.1% (95%: 97.5–98.6%) Reading 97.7% · grammar 98.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know”: Queued (its run with the option has not started yet) Confidence: calibration gap 7.5 points (written estimate) Cost $53.381$53,381 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 81.1 s median per request, 1 in flight | 98.1%97.5%–98.6% | 97.7% | 98.3% | Queued | 7.5written estimate | $53.381$53,381subscription, priced as the API | 81.116 per request1 in flight |
| 1 | Gemini 3.8 Flash high · chat‡ Closed · WorkhorseLatency 53.7 sCost $2.419$2,419 per 1,000per 1M Gemini 3.8 Flash high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.1% · grammar 98.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.5% · answered 99.9% · error rate 1.9% Confidence: calibration gap 1.1 points (written estimate) Cost $2.419$2,419 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 53.7 s median per request, 4 in flight | 98.1%97.4%–98.6% | 97.1% | 98.5% | 4.5%answered 100% | 1.1written estimate | $2.419$2,419at list prices | 53.716 per request4 in flight |
| Gemini 3.8 Flash low · chat‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.19 sCost $0.269$269 per 1,000per 1M Gemini 3.8 Flash low · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy 93.3% (95%: 92.1–94.2%) Reading 95.1% · grammar 92.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.9% · error rate 7.2% Confidence: calibration gap 3.5 points (written estimate) Cost $0.269$269 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 2.19 s median per request, 1 in flight | 93.3%92.1%–94.2% | 95.1% | 92.5% | 1.9%answered 100% | 3.5written estimate | $0.269$269at list prices | 2.1916 per request1 in flight | |
| Gemini 3.8 Flash low · chat · 1 per request‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.72 sCost $0.963$963 per 1,000per 1M Gemini 3.8 Flash low · chat · 1 per request · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy 97.1% (95%: 96.3–97.8%) Reading 96.2% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.2% · answered 99.8% · error rate 2.7% Confidence: calibration gap 0.6 points (written estimate) Cost $0.963$963 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.72 s median per request, 1 in flight | 97.1%96.3%–97.8% | 96.2% | 97.5% | 6.2%answered 100% | 0.6written estimate | $0.963$963at list prices | 2.721 per request1 in flight | |
| 1 | Gemini 4 Hydrogen high · chat‡ Closed · WorkhorseLatency 31.7 sCost $4.806$4,806 per 1,000per 1M Gemini 4 Hydrogen high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.3% · grammar 98.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.4% · answered 99.1% · error rate 1.7% Confidence: calibration gap 1.2 points (written estimate) Cost $4.806$4,806 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 31.7 s median per request, 1 in flight | 98.1%97.4%–98.6% | 97.3% | 98.4% | 17.4%answered 99% | 1.2written estimate | $4.806$4,806subscription, priced as the API | 31.716 per request1 in flight |
| 1 | GPT-6 Astra high · chat‡ Closed · FlagshipLatency 21.7 sCost $6.066$6,066 per 1,000per 1M GPT-6 Astra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 98.0% (95%: 97.3–98.5%) Reading 97.1% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.3% · answered 99.7% · error rate 1.9% Confidence: calibration gap 0.7 points (written estimate) Cost $6.066$6,066 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 21.7 s median per request, 1 in flight | 98.0%97.3%–98.5% | 97.1% | 98.3% | 14.3%answered 100% | 0.7written estimate | $6.066$6,066subscription, priced as the API | 21.716 per request1 in flight |
| GPT-6 Astra medium · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 15.3 sCost $5.447$5,447 per 1,000per 1M GPT-6 Astra medium · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 97.5% (95%: 96.8–98.1%) Reading 97.0% · grammar 97.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 12.7% · answered 99.7% · error rate 2.2% Confidence: calibration gap 1.0 points (written estimate) Cost $5.447$5,447 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 15.3 s median per request, 1 in flight | 97.5%96.8%–98.1% | 97.0% | 97.7% | 12.7%answered 100% | 1.0written estimate | $5.447$5,447subscription, priced as the API | 15.316 per request1 in flight | |
| GPT-6 Astra low · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 13.0 sCost $5.378$5,378 per 1,000per 1M GPT-6 Astra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 97.6% (95%: 96.9–98.2%) Reading 97.0% · grammar 97.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 10.5% · answered 99.7% · error rate 2.3% Confidence: calibration gap 0.8 points (written estimate) Cost $5.378$5,378 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.0 s median per request, 1 in flight | 97.6%96.9%–98.2% | 97.0% | 97.9% | 10.5%answered 100% | 0.8written estimate | $5.378$5,378subscription, priced as the API | 13.016 per request1 in flighton a sample | |
| GPT-6 Astra low · chat · 1 per request‡ Closed · Flagshipother setting of GPT-6 AstraLatency 6.41 sCost $26.738$26,738 per 1,000per 1M GPT-6 Astra low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 97.4% (95%: 96.7–98.0%) Reading 97.3% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.9% · answered 99.7% · error rate 2.4% Confidence: calibration gap 1.3 points (written estimate) Cost $26.738$26,738 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.41 s median per request, 1 in flight | 97.4%96.7%–98.0% | 97.3% | 97.5% | 11.9%answered 100% | 1.3written estimate | $26.738$26,738subscription, priced as the API | 6.411 per request1 in flight | |
| 1 | GPT-6.1 Sol high · chat‡ Closed · WorkhorseLatency 24.7 sCost $1.622$1,622 per 1,000per 1M GPT-6.1 Sol high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 97.8% (95%: 97.1–98.3%) Reading 96.5% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 15.4% · answered 99.6% · error rate 2.0% Confidence: calibration gap 1.0 points (written estimate) Cost $1.622$1,622 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 24.7 s median per request, 1 in flight | 97.8%97.1%–98.3% | 96.5% | 98.3% | 15.4%answered 100% | 1.0written estimate | $1.622$1,622subscription, priced as the API | 24.716 per request1 in flight |
| GPT-6.1 Sol low · chat‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 13.3 sCost $1.130$1,130 per 1,000per 1M GPT-6.1 Sol low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 96.7% (95%: 95.8–97.4%) Reading 96.2% · grammar 96.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 99.7% · error rate 2.7% Confidence: calibration gap 1.5 points (written estimate) Cost $1.130$1,130 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.3 s median per request, 1 in flight | 96.7%95.8%–97.4% | 96.2% | 96.9% | 9.2%answered 100% | 1.5written estimate | $1.130$1,130subscription, priced as the API | 13.316 per request1 in flight | |
| GPT-6.1 Sol low · chat · 1 per request‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 6.61 sCost $5.569$5,569 per 1,000per 1M GPT-6.1 Sol low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy 96.7% (95%: 95.9–97.4%) Reading 96.0% · grammar 97.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.6% · answered 99.6% · error rate 2.8% Confidence: calibration gap 1.8 points (written estimate) Cost $5.569$5,569 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.61 s median per request, 1 in flight | 96.7%95.9%–97.4% | 96.0% | 97.0% | 11.6%answered 100% | 1.8written estimate | $5.569$5,569subscription, priced as the API | 6.611 per request1 in flight | |
| 2 | Gemini 4 Argon low · chat‡ Closed · FlagshipLatency 5.46 sCost $1.786$1,786 per 1,000per 1M Gemini 4 Argon low · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-28) Accuracy 96.4% (95%: 95.5–97.1%) Reading 97.0% · grammar 96.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.0% · answered 99.7% · error rate 3.6% Confidence: calibration gap 2.2 points (written estimate) Cost $1.786$1,786 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.46 s median per request, 1 in flight | 96.4%95.5%–97.1% | 97.0% | 96.1% | 7.0%answered 100% | 2.2written estimate | $1.786$1,786at list prices | 5.4616 per request1 in flight |
| 2 | Claude Fable 5.1 high · chat‡ Closed · FlagshipLatency 30.3 sCost $17.042$17,042 per 1,000per 1M Claude Fable 5.1 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-28) Accuracy 96.1% (95%: 95.2–96.9%) Reading 95.6% · grammar 96.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.0% · answered 99.8% · error rate 3.6% Confidence: calibration gap 12.0 points (written estimate) Cost $17.042$17,042 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 30.3 s median per request, 1 in flight | 96.1%95.2%–96.9% | 95.6% | 96.4% | 6.0%answered 100% | 12.0written estimate | $17.042$17,042subscription, priced as the API | 30.316 per request1 in flight |
| 2 | Claude Opus 5.5 high · chat‡ Closed · FlagshipLatency 22.1 sCost $6.509$6,509 per 1,000per 1M Claude Opus 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 95.0% (95%: 94.0–95.8%) Reading 95.6% · grammar 94.7% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 2.7% · answered 99.9% · error rate 4.9% Confidence: calibration gap 14.6 points (written estimate) Cost $6.509$6,509 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 22.1 s median per request, 1 in flight | 95.0%94.0%–95.8% | 95.6% | 94.7% | 2.7%answered 100% | 14.6written estimate | $6.509$6,509subscription, priced as the API | 22.116 per request1 in flight |
| Claude Opus 5.5 low · chat‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 13.4 sCost $4.931$4,931 per 1,000per 1M Claude Opus 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 90.4% (95%: 89.1–91.6%) Reading 94.8% · grammar 88.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.9% · error rate 7.7% Confidence: calibration gap 16.5 points (written estimate) Cost $4.931$4,931 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.4 s median per request, 1 in flight | 90.4%89.1%–91.6% | 94.8% | 88.6% | 1.8%answered 100% | 16.5written estimate | $4.931$4,931subscription, priced as the API | 13.416 per request1 in flighton a sample | |
| Claude Opus 5.5 low · chat · 1 per request‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 3.71 sCost $11.610$11,610 per 1,000per 1M Claude Opus 5.5 low · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 93.3% (95%: 92.2–94.3%) Reading 95.7% · grammar 92.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 100.0% · error rate 6.2% Confidence: calibration gap 10.2 points (written estimate) Cost $11.610$11,610 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 3.71 s median per request, 1 in flight | 93.3%92.2%–94.3% | 95.7% | 92.3% | 0.7%answered 100% | 10.2written estimate | $11.610$11,610subscription, priced as the API | 3.711 per request1 in flight | |
| 3 | GPT-6.1 Terra high · chat‡ Closed · WorkhorseLatency 7.10 sCost $0.515$515 per 1,000per 1M GPT-6.1 Terra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Accuracy 94.3% (95%: 93.2–95.2%) Reading 96.2% · grammar 93.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.3% · answered 99.6% · error rate 5.3% Confidence: calibration gap 2.8 points (written estimate) Cost $0.515$515 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.10 s median per request, 1 in flight | 94.3%93.2%–95.2% | 96.2% | 93.5% | 7.3%answered 100% | 2.8written estimate | $0.515$515subscription, priced as the API | 7.1016 per request1 in flight |
| GPT-6.1 Terra low · chat‡ Closed · Workhorseother setting of GPT-6.1 TerraLatency 4.24 sCost $0.426$426 per 1,000per 1M GPT-6.1 Terra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Accuracy 90.9% (95%: 89.6–92.0%) Reading 95.9% · grammar 88.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.9% · answered 99.7% · error rate 9.1% Confidence: calibration gap 6.0 points (written estimate) Cost $0.426$426 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 4.24 s median per request, 1 in flight | 90.9%89.6%–92.0% | 95.9% | 88.7% | 2.9%answered 100% | 6.0written estimate | $0.426$426subscription, priced as the API | 4.2416 per request1 in flight | |
| 3 | Grok 4.7 high · chat‡ Closed · WorkhorseLatency 401.9 sCost $14.292$14,292 per 1,000per 1M Grok 4.7 high · chat · xAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://docs.x.ai/docs/models/grok-4.7, read 2026-10-28) Accuracy 92.7% (95%: 91.6–93.7%) Reading 95.0% · grammar 91.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.9% · answered 97.5% · error rate 6.1% Confidence: calibration gap 11.9 points (written estimate) Cost $14.292$14,292 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 401.9 s median per request, 4 in flight | 92.7%91.6%–93.7% | 95.0% | 91.8% | 29.9%answered 97% | 11.9written estimate | $14.292$14,292subscription, priced as the API | 401.916 per request4 in flight |
| 3 | DeepSeek-V4.1-Flash reasoning on · chat · first release‡ Open · 552B MoE · 16B activeLatency 365.1 sCost $1.299$1,299 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy 92.0% (95%: 90.8–93.1%) Reading 93.4% · grammar 91.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.0 points (answer letter probability) Cost $1.299$1,299 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 365.1 s median per request, 8 in flight With its dated checkpoint deepseek-v4.1-flash-1017 under its other settings | 92.0%90.8%–93.1% | 93.4% | 91.4% | Not run | 8.0answer letter probability | $1.299$1,299at a host's list price for the same weights | 365.116 per request8 in flight |
| DeepSeek-V4.1-Flash reasoning on · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 486.9 sCost $1.278$1,278 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy 91.8% (95%: 90.5–92.8%) Reading 94.1% · grammar 90.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.9% · error rate 7.0% Confidence: calibration gap 6.9 points (answer letter probability) Cost $1.278$1,278 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 486.9 s median per request, 8 in flight | 91.8%90.5%–92.8% | 94.1% | 90.8% | 1.3%answered 100% | 6.9answer letter probability | $1.278$1,278at the maker's list price | 486.916 per request8 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · first release‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 4.36 sCost $0.098$98.46 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy 65.9% (95%: 63.9–67.9%) Reading 87.8% · grammar 56.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.098$98.46 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 4.36 s median per request, 1 in flight | 65.9%63.9%–67.9% | 87.8% | 56.6% | Not run | 13.5answer letter probability | $0.098$98.46at a host's list price for the same weights | 4.3616 per request1 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 5.08 sCost $0.098$98.47 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy 66.0% (95%: 64.0–68.0%) Reading 88.3% · grammar 56.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 33.2% Confidence: calibration gap 16.4 points (answer letter probability) Cost $0.098$98.47 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 5.08 s median per request, 1 in flight | 66.0%64.0%–68.0% | 88.3% | 56.5% | 0.0%answered 100% | 16.4answer letter probability | $0.098$98.47at the maker's list price | 5.0816 per request1 in flight | |
| 4 | Qwen4-Max high · chat‡ Closed · WorkhorseLatency 91.5 sCost $2.533$2,533 per 1,000per 1M Qwen4-Max high · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://openrouter.ai/qwen/qwen4-max, read 2026-11-16) Accuracy 91.7% (95%: 90.5–92.8%) Reading 94.7% · grammar 90.5% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.2% · error rate 6.4% Confidence: calibration gap 7.8 points (written estimate) Cost $2.533$2,533 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 91.5 s median per request, 4 in flight | 91.7%90.5%–92.8% | 94.7% | 90.5% | 0.7%answered 99% | 7.8written estimate | $2.533$2,533at list prices | 91.516 per request4 in flight |
| 5 | Muse Ember max · chat‡ Closed · EfficientLatency 30.9 sCost $0.650$650 per 1,000per 1M Muse Ember max · chat · Meta · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/meta/muse-ember, read 2026-11-10) Accuracy 85.6% (95%: 84.1–87.0%) Reading 90.2% · grammar 83.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.1% · answered 97.5% · error rate 13.1% Confidence: calibration gap 6.2 points (written estimate) Cost $0.650$650 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 30.9 s median per request, 4 in flight | 85.6%84.1%–87.0% | 90.2% | 83.7% | 2.1%answered 98% | 6.2written estimate | $0.650$650at list prices | 30.916 per request4 in flight |
| 5 | Claude Sonnet 5.5 high · chat‡ Closed · WorkhorseLatency 20.5 sCost $3.654$3,654 per 1,000per 1M Claude Sonnet 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 85.0% (95%: 83.5–86.5%) Reading 85.2% · grammar 85.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.9% · error rate 12.8% Confidence: calibration gap 16.0 points (written estimate) Cost $3.654$3,654 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 20.5 s median per request, 1 in flight | 85.0%83.5%–86.5% | 85.2% | 85.0% | 0.7%answered 100% | 16.0written estimate | $3.654$3,654subscription, priced as the API | 20.516 per request1 in flighton a sample |
| Claude Sonnet 5.5 high · chat · 1 per request‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 4.64 sCost $7.748$7,748 per 1,000per 1M Claude Sonnet 5.5 high · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 89.2% (95%: 87.8–90.4%) Reading 94.2% · grammar 87.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.9% · error rate 11.1% Confidence: calibration gap 9.8 points (written estimate) Cost $7.748$7,748 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 4.64 s median per request, 1 in flight | 89.2%87.8%–90.4% | 94.2% | 87.0% | 0.8%answered 100% | 9.8written estimate | $7.748$7,748subscription, priced as the API | 4.641 per request1 in flight | |
| Claude Sonnet 5.5 low · chat‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 3.87 sCost $1.896$1,896 per 1,000per 1M Claude Sonnet 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy 61.4% (95%: 59.3–63.4%) Reading 82.9% · grammar 52.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.8% · error rate 35.1% Confidence: calibration gap 7.9 points (written estimate) Cost $1.896$1,896 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 3.87 s median per request, 1 in flight | 61.4%59.3%–63.4% | 82.9% | 52.2% | 0.6%answered 100% | 7.9written estimate | $1.896$1,896subscription, priced as the API | 3.8716 per request1 in flighton a sample | |
| 5 | Kimi K3.1 Pro low · chat‡ Open · 2.8T MoE · 104B activeLatency 44.9 sCost $2.048$2,048 per 1,000per 1M Kimi K3.1 Pro low · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Accuracy 82.8% (95%: 81.1–84.3%) Reading 93.8% · grammar 78.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.0% · answered 99.7% · error rate 15.7% Confidence: calibration gap 12.6 points (written estimate) Cost $2.048$2,048 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 44.9 s median per request, 4 in flight | 82.8%81.1%–84.3% | 93.8% | 78.1% | 2.0%answered 100% | 12.6written estimate | $2.048$2,048at list prices | 44.916 per request4 in flight |
| Kimi K3.1 Pro no reasoning · chat‡ Open · 2.8T MoE · 104B activeother setting of Kimi K3.1 ProLatency 5.75 sCost $0.853$853 per 1,000per 1M Kimi K3.1 Pro no reasoning · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Accuracy 72.7% (95%: 70.8–74.5%) Reading 92.5% · grammar 64.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.9% · error rate 28.3% Confidence: calibration gap 20.1 points (written estimate) Cost $0.853$853 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.75 s median per request, 4 in flight | 72.7%70.8%–74.5% | 92.5% | 64.3% | 0.5%answered 100% | 20.1written estimate | $0.853$853at list prices | 5.7516 per request4 in flight | |
| 5 | Qwen3.5-397B-A17B high · chat‡ Open · 397B MoE · 17B activeLatency 251.3 sCost $2.411$2,411 per 1,000per 1M Qwen3.5-397B-A17B high · chat · Alibaba · chat model Mixture of experts: 397B total, about 17B active per step. Size class uses the total (397B). Accuracy 82.7% (95%: 81.1–84.2%) Reading 91.0% · grammar 79.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.5% · answered 99.7% · error rate 17.8% Confidence: calibration gap 6.2 points (written estimate) Cost $2.411$2,411 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 251.3 s median per request, 4 in flight | 82.7%81.1%–84.2% | 91.0% | 79.2% | 1.5%answered 100% | 6.2written estimate | $2.411$2,411at list prices | 251.316 per request4 in flight |
| 6 | KnowLine-31B-Gen1.5‡ Open · 31BLatency 1.05 sCost No list price KnowLine-31B-Gen1.5 · PelaAI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 81.8% (95%: 80.1–83.4%) Reading 92.8% · grammar 77.1% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.2% · answered 100.0% · error rate 18.7% Kept at 99% confidence with no observed mistakes: 6.5% Confidence: calibration gap 3.4 points (option score) Cost: No list price · 1 question per request Latency: 1.05 s median per request, 1 in flight | 81.8%80.1%–83.4% | 92.8% | 77.1% | 0.2%answered 100% | 3.4option score | No list price | 1.051 per request1 in flight |
| 6 | Copernicus-31B‡ Open · 31BLatency 0.184 sCost $0.134$134 per 1,000per 1M Copernicus-31B · Poindexter Labs · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 81.8% (95%: 80.1–83.4%) Reading 93.1% · grammar 77.0% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.2% · answered 99.8% · error rate 18.3% Kept at 99% confidence with no observed mistakes: 4.9% Confidence: calibration gap 9.1 points (option score) Cost $0.134$134 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.184 s median per request, 1 in flight | 81.8%80.1%–83.4% | 93.1% | 77.0% | 1.2%answered 100% | 9.1option score | $0.134$134from GPU time | 0.1841 per request1 in flight |
| 6 | KnowLine-31B-Gen2‡ Open · 31BLatency 0.168 sCost $0.120$120 per 1,000per 1M KnowLine-31B-Gen2 · PelaAI · decision model Base model: Gemma 4 31B Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 81.2% (95%: 79.5–82.8%) Reading 93.4% · grammar 76.0% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 18.6% Kept at 99% confidence with no observed mistakes: 3.7% Confidence: calibration gap 5.6 points (option score) Cost $0.120$120 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.168 s median per request, 1 in flight | 81.2%79.5%–82.8% | 93.4% | 76.0% | 0.0%answered 100% | 5.6option score | $0.120$120from GPU time | 0.1681 per request1 in flight |
| 6 | Turion Jev 31B‡ Open · 31BLatency 0.405 sCost No list price Turion Jev 31B · Turion.AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 80.1% (95%: 78.4–81.7%) Reading 92.7% · grammar 74.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.3% · answered 99.3% · error rate 20.0% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 11.1 points (option score) Cost: No list price · 1 question per request Latency: 0.405 s median per request, 1 in flight | 80.1%78.4%–81.7% | 92.7% | 74.8% | 3.3%answered 99% | 11.1option score | No list price | 0.4051 per request1 in flight |
| 6 | Inkling-Small high · chat · Jul 2026‡ Open · 276B MoE · 12B activeLatency 97.7 sCost $1.137$1,137 per 1,000per 1M Inkling-Small high · chat · Jul 2026 · Thinking Machines · chat model Mixture of experts: 276B total (counted from the released files: 265.96B), about 12B active per step. Size class uses the total (276B). Accuracy 80.0% (95%: 78.3–81.6%) Reading 87.2% · grammar 76.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.1% · answered 99.3% · error rate 21.4% Confidence: calibration gap 7.9 points (written estimate) Cost $1.137$1,137 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 97.7 s median per request, 4 in flight | 80.0%78.3%–81.6% | 87.2% | 76.9% | 3.1%answered 99% | 7.9written estimate | $1.137$1,137at list prices | 97.716 per request4 in flight |
| 6 | GLM-5.3-Flash reasoning on · chat‡ Open · 320B MoE · 18B activeLatency 102.6 sCost $0.543$543 per 1,000per 1M GLM-5.3-Flash reasoning on · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Accuracy 78.8% (95%: 77.0–80.5%) Reading 89.3% · grammar 74.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 91.9% · error rate 16.6% Confidence: calibration gap 4.5 points (written estimate) Cost $0.543$543 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 102.6 s median per request, 8 in flight | 78.8%77.0%–80.5% | 89.3% | 74.3% | 0.6%answered 92% | 4.5written estimate | $0.543$543at the maker's list price | 102.616 per request8 in flight |
| GLM-5.3-Flash reasoning off · chat‡ Open · 320B MoE · 18B activeother setting of GLM-5.3-FlashLatency 1.07 sCost $0.047$47.28 per 1,000per 1M GLM-5.3-Flash reasoning off · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Accuracy 57.8% (95%: 55.7–59.8%) Reading 82.2% · grammar 47.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 41.0% Confidence: calibration gap 21.0 points (written estimate) Cost $0.047$47.28 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 1.07 s median per request, 1 in flight | 57.8%55.7%–59.8% | 82.2% | 47.4% | 0.3%answered 100% | 21.0written estimate | $0.047$47.28at the maker's list price | 1.0716 per request1 in flight | |
| 7 | REAX 1.3‡ Open · 57BLatency 0.250 sCost $0.185$185 per 1,000per 1M REAX 1.3 · REAX · decision model Two models run together, option probabilities averaged 50/50: 57.36B in all, all used on every question. Accuracy 77.9% (95%: 76.2–79.6%) Reading 92.4% · grammar 71.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 4.1 points (option score) Cost $0.185$185 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 77.9%76.2%–79.6% | 92.4% | 71.8% | 0.4%answered 100% | 4.1option score | $0.185$185from GPU time | 0.2501 per request1 in flight |
| 7 | Karpo Decision 31B‡ Open · 31BLatency 0.159 sCost $0.105$105 per 1,000per 1M Karpo Decision 31B · MachinePulse · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 77.8% (95%: 76.0–79.5%) Reading 93.4% · grammar 71.1% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 23.0% Confidence: calibration gap 2.3 points (option score) Cost $0.105$105 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.159 s median per request, 1 in flight | 77.8%76.0%–79.5% | 93.4% | 71.1% | 0.6%answered 100% | 2.3option score | $0.105$105from GPU time | 0.1591 per request1 in flight |
| 7 | Gemini Decision 1 20261111‡ Closed · EfficientLatency 0.227 sCost $0.037$37.19 per 1,000per 1M Gemini Decision 1 20261111 · Google · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/google/gemini-decision-1, read 2026-11-11) Accuracy 77.5% (95%: 75.7–79.2%) Reading 92.1% · grammar 71.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.5% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 14.5% Confidence: calibration gap 2.0 points (option score) Cost $0.037$37.19 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.227 s median per request, 1 in flight | 77.5%75.7%–79.2% | 92.1% | 71.3% | 1.9%answered 100% | 2.0option score | $0.037$37.19at list prices | 0.2271 per request1 in flight |
| 7 | d1-ultra‡ Closed · EfficientLatency 0.223 sCost $0.014$14.34 per 1,000per 1M d1-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-ultra, read 2026-10-16) Accuracy 77.1% (95%: 75.3–78.8%) Reading 92.1% · grammar 70.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 75.1% · answered 71.9% · error rate 13.0% Kept at 99% confidence with no observed mistakes: 4.1% Confidence: calibration gap 3.6 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.223 s median per request, 1 in flight | 77.1%75.3%–78.8% | 92.1% | 70.7% | 75.1%answered 72% | 3.6option score | $0.014$14.34at list prices | 0.2231 per request1 in flight |
| 8 | Logits Sigma-1 Duo‡ Open · 57BLatency 0.250 sCost $0.181$181 per 1,000per 1M Logits Sigma-1 Duo · Logits · decision model Parameters: 57.36B dense, as the maker states it. Accuracy 75.0% (95%: 73.2–76.8%) Reading 91.9% · grammar 67.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 24.7% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 5.7 points (option score) Cost $0.181$181 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 75.0%73.2%–76.8% | 91.9% | 67.8% | 0.4%answered 100% | 5.7option score | $0.181$181from GPU time | 0.2501 per request1 in flight |
| 8 | Qwen3.8-27B low · chat‡ Open · 27BLatency 648.4 sCost $3.095$3,095 per 1,000per 1M Qwen3.8-27B low · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy 74.7% (95%: 72.8–76.4%) Reading 89.3% · grammar 68.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.9% · answered 97.9% · error rate 22.9% Confidence: calibration gap 23.7 points (answer letter probability) Cost $3.095$3,095 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 648.4 s median per request, 4 in flight | 74.7%72.8%–76.4% | 89.3% | 68.4% | 5.9%answered 98% | 23.7answer letter probability | $3.095$3,095at the maker's list price | 648.416 per request4 in flight |
| Qwen3.8-27B yes/no scoring‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.280 sCost $0.704$704 per 1,000per 1M Qwen3.8-27B yes/no scoring · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy 51.9% (95%: 49.8–54.0%) Reading 80.5% · grammar 39.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.9 points (option score) Cost $0.704$704 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.280 s median per request, 1 in flight | 51.9%49.8%–54.0% | 80.5% | 39.8% | Not run | 8.9option score | $0.704$704at the maker's list price | 0.2801 per request1 in flighton its repeats | |
| Qwen3.8-27B stated confidence · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.433 sCost $0.234$234 per 1,000per 1M Qwen3.8-27B stated confidence · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy 54.2% (95%: 52.1–56.3%) Reading 83.4% · grammar 41.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.7% · error rate 45.6% Confidence: calibration gap 40.8 points (written estimate) Cost $0.234$234 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.433 s median per request, 1 in flight | 54.2%52.1%–56.3% | 83.4% | 41.8% | 0.6%answered 100% | 40.8written estimate | $0.234$234at the maker's list price | 0.4331 per request1 in flighton its repeats | |
| Qwen3.8-27B vote share · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 1.62 sCost $0.825$825 per 1,000per 1M Qwen3.8-27B vote share · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy 54.3% (95%: 52.2–56.4%) Reading 84.0% · grammar 41.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.4% · error rate 45.6% Confidence: calibration gap 22.5 points (agreement across ten answers) Cost $0.825$825 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 1.62 s median per request, 1 in flight | 54.3%52.2%–56.4% | 84.0% | 41.7% | 1.3%answered 99% | 22.5agreement across ten answers | $0.825$825at the maker's list price | 1.621 per request1 in flight | |
| 8 | GLM-5.3 low · chat‡ Open · WorkhorseLatency 17.1 sCost $0.814$814 per 1,000per 1M GLM-5.3 low · chat · Z.ai · chat model Parameters: not stated by the maker (a mixture of experts); counted from the released files: 753.33B. Class: Workhorse, by its standard list price: $1.40 per 1M input tokens (https://openrouter.ai/z-ai/glm-5.3, read 2026-10-28) Accuracy 74.5% (95%: 72.7–76.3%) Reading 87.7% · grammar 68.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.1% · answered 99.0% · error rate 23.9% Confidence: calibration gap 2.6 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 17.1 s median per request, 4 in flight | 74.5%72.7%–76.3% | 87.7% | 68.9% | 1.1%answered 99% | 2.6written estimate | $0.814$814at list prices | 17.116 per request4 in flight |
| 8 | Sieve 31B v3‡ Open · 31BLatency 1.00 sCost No list price Sieve 31B v3 · Sthānika AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 74.2% (95%: 72.3–75.9%) Reading 91.8% · grammar 66.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 25.2% Kept at 99% confidence: 15.0%, 1 of them wrong Confidence: calibration gap 2.9 points (option score) Cost: No list price · 1 question per request Latency: 1.00 s median per request, 1 in flight | 74.2%72.3%–75.9% | 91.8% | 66.7% | 0.0%answered 100% | 2.9option score | No list price | 1.001 per request1 in flight |
| 8 | Gemma 4 31B reasoning off · chat‡ Open · 31BLatency 3.97 sCost $0.048$48.22 per 1,000per 1M Gemma 4 31B reasoning off · chat · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 73.1% (95%: 71.2–74.9%) Reading 90.9% · grammar 65.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.1% · answered 98.9% · error rate 26.1% Confidence: calibration gap 24.2 points (answer letter probability) Cost $0.048$48.22 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 3.97 s median per request, 1 in flight | 73.1%71.2%–74.9% | 90.9% | 65.6% | 4.1%answered 99% | 24.2answer letter probability | $0.048$48.22at a host's list price for the same weights | 3.9716 per request1 in flight |
| Gemma 4 31B reasoning off · chat · 1 per request‡ Open · 31Bother setting of Gemma 4 31BLatency 0.298 sCost $0.068$68.19 per 1,000per 1M Gemma 4 31B reasoning off · chat · 1 per request · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Accuracy 66.8% (95%: 64.8–68.7%) Reading 89.0% · grammar 57.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.4% · error rate 33.5% Confidence: calibration gap 31.2 points (answer letter probability) Cost $0.068$68.19 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.298 s median per request, 1 in flight | 66.8%64.8%–68.7% | 89.0% | 57.3% | 1.9%answered 99% | 31.2answer letter probability | $0.068$68.19at a host's list price for the same weights | 0.2981 per request1 in flight | |
| 8 | MiMo-V2.6-Flash high · chat‡ Open · 309B MoE · 15B activeLatency 21.4 sCost $2.219$2,219 per 1,000per 1M MiMo-V2.6-Flash high · chat · Xiaomi · chat model Mixture of experts: 309B total, about 15B active per step. Size class uses the total (309B). Accuracy 72.1% (95%: 70.2–73.9%) Reading 86.6% · grammar 66.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.5% · answered 96.7% · error rate 26.2% Confidence: calibration gap 7.1 points (written estimate) Cost $2.219$2,219 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 21.4 s median per request, 32 in flight | 72.1%70.2%–73.9% | 86.6% | 66.0% | 3.5%answered 97% | 7.1written estimate | $2.219$2,219at list prices | 21.41 per request32 in flight |
| 9 | GLiDE 27B‡ Open · 27BLatency 0.542 sCost No list price GLiDE 27B · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 71.1% (95%: 69.2–73.0%) Reading 90.9% · grammar 62.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.8% · answered 86.8% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 11.2% Confidence: calibration gap 2.3 points (option score) Cost: No list price · 1 question per request Latency: 0.542 s median per request, 1 in flight | 71.1%69.2%–73.0% | 90.9% | 62.7% | 39.8%answered 87% | 2.3option score | No list price | 0.5421 per request1 in flight |
| – | Perplexity Decider 27B v1.2 API · with web‡ Open · 27BLatency 0.482 sCost ≤ $1.034≤ $1,034 per 1,000per 1M Perplexity Decider 27B v1.2 API · with web · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 68.7% (95%: 66.8–70.6%) Reading 88.9% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 99.0% · error rate 32.3% Kept at 99% confidence: 16.0%, 1 of them wrong Confidence: calibration gap 3.6 points (option score) Cost ≤ $1.034≤ $1,034 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.482 s median per request, 1 in flight | 68.7%66.8%–70.6% | 88.9% | 60.2% | 3.2%answered 99% | 3.6option score | ≤ $1.034≤ $1,034at list prices | 0.4821 per request1 in flight |
| 9 | GLiDE v1.2‡ Open · 27BLatency 0.413 sCost $0.014$14.37 per 1,000per 1M GLiDE v1.2 · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 68.5% (95%: 66.5–70.4%) Reading 89.6% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.5% · answered 85.4% · error rate 26.2% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 2.9 points (option score) Cost $0.014$14.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.413 s median per request, 1 in flight | 68.5%66.5%–70.4% | 89.6% | 59.5% | 39.5%answered 85% | 2.9option score | $0.014$14.37at the maker's list price | 0.4131 per request1 in flight |
| 9 | Perplexity Decider 27B v1.3‡ Open · 27BLatency 0.110 sCost $0.081$81.18 per 1,000per 1M Perplexity Decider 27B v1.3 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 68.4% (95%: 66.5–70.3%) Reading 89.0% · grammar 59.7% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.8% · error rate 32.4% Kept at 99% confidence with no observed mistakes: 12.6% Confidence: calibration gap 5.0 points (option score) Cost $0.081$81.18 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.110 s median per request, 1 in flight | 68.4%66.5%–70.3% | 89.0% | 59.7% | 0.7%answered 100% | 5.0option score | $0.081$81.18from GPU time | 0.1101 per request1 in flight |
| 9 | GPT-6 Luna low · chat‡ Closed · EfficientLatency 7.71 sCost $0.048$48.01 per 1,000per 1M GPT-6 Luna low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Accuracy 68.3% (95%: 66.3–70.2%) Reading 87.3% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.0% · answered 99.7% · error rate 33.2% Confidence: calibration gap 25.7 points (written estimate) Cost $0.048$48.01 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.71 s median per request, 1 in flight | 68.3%66.3%–70.2% | 87.3% | 60.2% | 1.0%answered 100% | 25.7written estimate | $0.048$48.01subscription, priced as the API | 7.7116 per request1 in flight |
| 9 | d1.2-ultra‡ Closed · EfficientLatency 0.220 sCost $0.014$14.34 per 1,000per 1M d1.2-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Accuracy 68.1% (95%: 66.1–70.0%) Reading 89.8% · grammar 58.9% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 87.3% · answered 54.7% · error rate 12.1% Kept at 99% confidence with no observed mistakes: 0.5% Confidence: calibration gap 3.4 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.220 s median per request, 1 in flight | 68.1%66.1%–70.0% | 89.8% | 58.9% | 87.3%answered 55% | 3.4option score | $0.014$14.34at list prices | 0.2201 per request1 in flight |
| 9 | GLiDE API‡ Closed · EfficientLatency 2.09 sCost $0.160$160 per 1,000per 1M GLiDE API · Fastino · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.fastino.ai/pricing, read 2026-10-22) Accuracy 67.9% (95%: 65.9–69.8%) Reading 87.8% · grammar 59.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 59.4% · answered 76.3% · error rate 21.2% Kept at 99% confidence with no observed mistakes: 5.5% Confidence: calibration gap 4.5 points (option score) Cost $0.160$160 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.09 s median per request, 4 in flight | 67.9%65.9%–69.8% | 87.8% | 59.4% | 59.4%answered 76% | 4.5option score | $0.160$160at list prices | 2.091 per request4 in flight |
| 10 | d3‡ Open · 26BLatency 0.092 sCost $0.070$70.21 per 1,000per 1M d3 · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.2% · grammar 58.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 32.5% Kept at 99% confidence with no observed mistakes: 11.4% Confidence: calibration gap 2.6 points (option score) Cost $0.070$70.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.092 s median per request, 1 in flight | 67.8%65.8%–69.7% | 89.2% | 58.7% | 0.0%answered 100% | 2.6option score | $0.070$70.21from GPU time | 0.0921 per request1 in flight |
| 10 | d3-ultra‡ Open · 26BLatency 0.333 sCost $0.216$216 per 1,000per 1M d3-ultra · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.8% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.5% · error rate 32.2% Kept at 99% confidence: 17.5%, 1 of them wrong Confidence: calibration gap 12.5 points (option score) Cost $0.216$216 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.333 s median per request, 1 in flight | 67.8%65.8%–69.7% | 89.8% | 58.4% | 1.4%answered 100% | 12.5option score | $0.216$216from GPU time | 0.3331 per request1 in flight |
| 10 | Drex Max‡ Open · 27BLatency 0.082 sCost $0.023$23.29 per 1,000per 1M Drex Max · Nace.AI · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 67.7% (95%: 65.8–69.7%) Reading 89.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.4% Kept at 99% confidence: 20.2%, 1 of them wrong Confidence: calibration gap 6.3 points (option score) Cost $0.023$23.29 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.082 s median per request, 1 in flight | 67.7%65.8%–69.7% | 89.6% | 58.4% | 0.3%answered 100% | 6.3option score | $0.023$23.29at the maker's list price | 0.0821 per request1 in flight |
| 10 | Drex Max v1.1‡ Closed · Efficient (provisional)Latency 0.313 sCost $0.023$23.11 per 1,000per 1M Drex Max v1.1 · Nace.AI · decision model Parameters: not disclosed by maker Class: Efficient (provisional), by its introductory price, the only one published: $0.03 per 1M input tokens (https://nace.ai/developers, read 2026-11-10) Accuracy 67.4% (95%: 65.4–69.3%) Reading 88.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.2% · answered 85.4% · error rate 27.5% Kept at 99% confidence with no observed mistakes: 10.0% Confidence: calibration gap 2.9 points (option score) Cost $0.023$23.11 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 67.4%65.4%–69.3% | 88.6% | 58.4% | 38.2%answered 85% | 2.9option score | $0.023$23.11at list prices | 0.3131 per request1 in flight |
| 10 | Mammoth-1‡ Open · 27BLatency 0.169 sCost $0.040$39.62 per 1,000per 1M Mammoth-1 · Base Labs · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Accuracy 67.1% (95%: 65.1–69.0%) Reading 90.4% · grammar 57.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 68.7% · answered 68.6% · error rate 20.8% Kept at 99% confidence with no observed mistakes: 13.0% Confidence: calibration gap 5.6 points (option score) Cost $0.040$39.62 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.169 s median per request, 1 in flight | 67.1%65.1%–69.0% | 90.4% | 57.2% | 68.7%answered 69% | 5.6option score | $0.040$39.62at the maker's list price | 0.1691 per request1 in flight |
| 10 | Perplexity Decider 27B v1.2 DGX Spark‡ Open · 27BLatency 0.580 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.2 DGX Spark · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 67.1% (95%: 65.1–69.0%) Reading 89.2% · grammar 57.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.7% Kept at 99% confidence with no observed mistakes: 16.4% Confidence: calibration gap 4.4 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.580 s median per request, 1 in flight | 67.1%65.1%–69.0% | 89.2% | 57.7% | 0.3%answered 100% | 4.4option score | $0.008$7.76at the maker's list price | 0.5801 per request1 in flight |
| Perplexity Decider 27B v1.2 API‡ Open · 27Bother setting of Perplexity Decider 27B v1.2Latency 0.254 sCost $0.008$7.70 per 1,000per 1M Perplexity Decider 27B v1.2 API · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 66.3% (95%: 64.3–68.3%) Reading 89.0% · grammar 56.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 33.1% Kept at 99% confidence with no observed mistakes: 14.9% Confidence: calibration gap 4.2 points (option score) Cost $0.008$7.70 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.254 s median per request, 1 in flight | 66.3%64.3%–68.3% | 89.0% | 56.7% | 0.4%answered 100% | 4.2option score | $0.008$7.70at list prices | 0.2541 per request1 in flight | |
| 10 | Perplexity Decider 27B v1.1‡ Open · 27BLatency 0.233 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 66.9% (95%: 64.9–68.8%) Reading 89.0% · grammar 57.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 15.9% Confidence: calibration gap 3.8 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.233 s median per request, 1 in flight | 66.9%64.9%–68.8% | 89.0% | 57.5% | Not run | 3.8option score | $0.008$7.76at list prices | 0.2331 per request1 in flight |
| 10 | MiniMax-M3.1-Flash high · chat‡ Open · 428B MoE · 23B activeLatency 112.6 sCost $0.814$814 per 1,000per 1M MiniMax-M3.1-Flash high · chat · MiniMax · chat model Mixture of experts: 428B total, about 23B active per step. Size class uses the total (428B). Accuracy 66.5% (95%: 64.5–68.4%) Reading 82.9% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.5% · answered 97.6% · error rate 32.8% Confidence: calibration gap 21.1 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 112.6 s median per request, 4 in flight | 66.5%64.5%–68.4% | 82.9% | 59.5% | 5.5%answered 98% | 21.1written estimate | $0.814$814at list prices | 112.616 per request4 in flight |
| 10 | SubQ Snap Preview‡ Closed · EfficientLatency 0.389 sCost $0.019$19.41 per 1,000per 1M SubQ Snap Preview · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.05 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-10-13) Accuracy 65.2% (95%: 63.2–67.2%) Reading 87.7% · grammar 55.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.9 points (option score) Cost $0.019$19.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.389 s median per request, 1 in flight | 65.2%63.2%–67.2% | 87.7% | 55.7% | Not run | 4.9option score | $0.019$19.41at list prices | 0.3891 per request1 in flight |
| 11 | SubQ Snap 1.1‡ Closed · EfficientLatency 0.378 sCost $0.015$15.41 per 1,000per 1M SubQ Snap 1.1 · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-11-01) Accuracy 63.9% (95%: 61.8–65.9%) Reading 87.0% · grammar 54.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 47.1% · answered 80.1% · error rate 27.8% Kept at 99% confidence: 13.6%, 1 of them wrong Confidence: calibration gap 11.2 points (option score) Cost $0.015$15.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.378 s median per request, 1 in flight | 63.9%61.8%–65.9% | 87.0% | 54.0% | 47.1%answered 80% | 11.2option score | $0.015$15.41at list prices | 0.3781 per request1 in flight |
| 11 | GPT-6 Luna Decisions API‡ Closed · EfficientLatency 0.191 sCost $0.040$40.44 per 1,000per 1M GPT-6 Luna Decisions API · OpenAI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Accuracy 63.6% (95%: 61.6–65.6%) Reading 85.5% · grammar 54.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 16.7%, 3 of them wrong Confidence: calibration gap 2.4 points (option score) Cost $0.040$40.44 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.191 s median per request, 4 in flight | 63.6%61.6%–65.6% | 85.5% | 54.3% | Not run | 2.4option score | $0.040$40.44at list prices | 0.1911 per request4 in flight |
| 11 | Qwen3.8 Flash no reasoning · chat‡ Closed · EfficientLatency 10.0 sCost $0.047$47.13 per 1,000per 1M Qwen3.8 Flash no reasoning · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.15 per 1M input tokens (https://openrouter.ai/qwen/qwen3.8-flash, read 2026-11-11) Accuracy 63.1% (95%: 61.0–65.1%) Reading 87.2% · grammar 52.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 98.3% · error rate 35.4% Confidence: calibration gap 28.7 points (written estimate) Cost $0.047$47.13 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 10.0 s median per request, 4 in flight | 63.1%61.0%–65.1% | 87.2% | 52.8% | 4.7%answered 98% | 28.7written estimate | $0.047$47.13at list prices | 10.016 per request4 in flight |
| 11 | KnowLine-12B-Gen1‡ Open · 12BLatency 0.071 sCost $0.052$52.28 per 1,000per 1M KnowLine-12B-Gen1 · PelaAI · decision model Base model: Gemma 4 12B Unified Parameters: 12B dense, as the maker states it (counted from the released files: 11.96B). Accuracy 62.5% (95%: 60.4–64.5%) Reading 86.0% · grammar 52.5% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 36.6% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.2 points (option score) Cost $0.052$52.28 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.071 s median per request, 1 in flight | 62.5%60.4%–64.5% | 86.0% | 52.5% | 0.1%answered 100% | 6.2option score | $0.052$52.28from GPU time | 0.0711 per request1 in flight |
| 11 | d1.2‡ Closed · EfficientLatency 0.213 sCost $0.007$6.83 per 1,000per 1M d1.2 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Accuracy 61.7% (95%: 59.7–63.7%) Reading 86.0% · grammar 51.4% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 79.4% · answered 55.7% · error rate 20.7% Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 3.3 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.213 s median per request, 1 in flight | 61.7%59.7%–63.7% | 86.0% | 51.4% | 79.4%answered 56% | 3.3option score | $0.007$6.83at list prices | 0.2131 per request1 in flight |
| 11 | d1.1‡ Closed · EfficientLatency 0.195 sCost $0.007$6.83 per 1,000per 1M d1.1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-24) Accuracy 61.5% (95%: 59.5–63.5%) Reading 84.6% · grammar 51.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 4.0% Confidence: calibration gap 2.0 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.195 s median per request, 1 in flight | 61.5%59.5%–63.5% | 84.6% | 51.7% | Not run | 2.0option score | $0.007$6.83at list prices | 0.1951 per request1 in flight |
| 11 | celeris-1-decision‡ Closed · EfficientLatency 0.401 sCost $0.015$15.28 per 1,000per 1M celeris-1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-10-10) Accuracy 60.5% (95%: 58.4–62.5%) Reading 85.2% · grammar 49.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.3 points (option score) Cost $0.015$15.28 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.401 s median per request, 1 in flight | 60.5%58.4%–62.5% | 85.2% | 49.9% | Not run | 4.3option score | $0.015$15.28at list prices | 0.4011 per request1 in flight |
| 11 | Perplexity Decider 27B v1‡ Open · 27BLatency 0.216 sCost $0.016$15.54 per 1,000per 1M Perplexity Decider 27B v1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy 60.4% (95%: 58.3–62.4%) Reading 86.0% · grammar 49.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.0% Confidence: calibration gap 5.0 points (option score) Cost $0.016$15.54 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.216 s median per request, 1 in flight | 60.4%58.3%–62.4% | 86.0% | 49.5% | Not run | 5.0option score | $0.016$15.54at list prices | 0.2161 per request1 in flighton its repeats |
| 12 | OneJev-27B‡ Open · 27BLatency 0.125 sCost $0.099$99.09 per 1,000per 1M OneJev-27B · OmniJev · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Accuracy 59.6% (95%: 57.5–61.6%) Reading 85.2% · grammar 48.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 40.6% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 5.5 points (option score) Cost $0.099$99.09 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.125 s median per request, 1 in flight | 59.6%57.5%–61.6% | 85.2% | 48.7% | 0.1%answered 100% | 5.5option score | $0.099$99.09from GPU time | 0.1251 per request1 in flight |
| 12 | Jev 1.14‡ Closed · EfficientLatency 0.211 sCost $0.029$28.84 per 1,000per 1M Jev 1.14 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://docs.typesafe.ai/models, read 2026-10-16) Accuracy 59.2% (95%: 57.1–61.2%) Reading 85.1% · grammar 48.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 11.8%, 1 of them wrong Confidence: calibration gap 2.5 points (option score) Cost $0.029$28.84 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.211 s median per request, 1 in flight | 59.2%57.1%–61.2% | 85.1% | 48.2% | Not run | 2.5option score | $0.029$28.84at list prices | 0.2111 per request1 in flight |
| 12 | Qwen3.5-4B reasoning on · chat‡ Open · 4.0BLatency 140.3 sCost $2.998$2,998 per 1,000per 1M Qwen3.5-4B reasoning on · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy 58.9% (95%: 56.8–61.0%) Reading 79.1% · grammar 50.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.3% · answered 98.4% · error rate 40.7% Confidence: not measured Cost $2.998$2,998 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 140.3 s median per request, 32 in flight | 58.9%56.8%–61.0% | 79.1% | 50.3% | 3.3%answered 98% | Not measureda plain answer line, no confidence | $2.998$2,998from GPU time | 140.31 per request32 in flight |
| 12 | Mercury Decide‡ Closed · EfficientLatency 0.313 sCost $0.008$7.71 per 1,000per 1M Mercury Decide · Inception · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://www.inceptionlabs.ai/, read 2026-11-10) Accuracy 58.3% (95%: 56.2–60.3%) Reading 84.5% · grammar 47.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 20.5%, 13 of them wrong Confidence: calibration gap 15.5 points (option score) Cost $0.008$7.71 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 58.3%56.2%–60.3% | 84.5% | 47.1% | Not run | 15.5option score | $0.008$7.71at list prices | 0.3131 per request1 in flight |
| 12 | Gemma 4 E4B default reasoning · chat‡ Open · 8.0BLatency 13.4 sCost $0.307$307 per 1,000per 1M Gemma 4 E4B default reasoning · chat · Google · chat model Parameters: 8B dense · 4.5B effective, as the maker states it (counted from the released files: 8.00B). Accuracy 57.6% (95%: 55.5–59.6%) Reading 77.6% · grammar 49.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.7% · answered 86.0% · error rate 38.4% Confidence: not measured Cost $0.307$307 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 13.4 s median per request, 32 in flight | 57.6%55.5%–59.6% | 77.6% | 49.0% | 29.7%answered 86% | Not measureda plain answer line, no confidence | $0.307$307from GPU time | 13.41 per request32 in flight |
| 12 | Jev 1.13‡ Closed · EfficientLatency 0.243 sCost $0.029$28.85 per 1,000per 1M Jev 1.13 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://typesafe.ai/blog/introducing-system-one-models-and-jev, read 2026-10-08) Accuracy 57.5% (95%: 55.4–59.6%) Reading 83.7% · grammar 46.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 14.2%, 1 of them wrong Confidence: calibration gap 5.1 points (option score) Cost $0.029$28.85 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.243 s median per request, 1 in flight | 57.5%55.4%–59.6% | 83.7% | 46.4% | Not run | 5.1option score | $0.029$28.85at list prices | 0.2431 per request1 in flighton its repeats |
| 12 | Clef‡ Open · 27BLatency 0.205 sCost $0.110$110 per 1,000per 1M Clef · Cloudflare · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.48B). Accuracy 57.3% (95%: 55.2–59.3%) Reading 85.1% · grammar 45.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.4% · answered 96.8% · error rate 42.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.7 points (option score) Cost $0.110$110 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.205 s median per request, 1 in flight | 57.3%55.2%–59.3% | 85.1% | 45.5% | 7.4%answered 97% | 10.7option score | $0.110$110at the maker's list price | 0.2051 per request1 in flight |
| 12 | Claude Haiku 5.5 max · chat‡ Closed · EfficientLatency 417.0 sCost $2.965$2,965 per 1,000per 1M Claude Haiku 5.5 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/anthropic/claude-haiku-5.5, read 2026-11-10) Accuracy 57.1% (95%: 55.0–59.1%) Reading 86.9% · grammar 44.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.0% · answered 67.7% · error rate 9.3% Confidence: calibration gap 17.5 points (written estimate) Cost $2.965$2,965 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 417.0 s median per request, 4 in flight | 57.1%55.0%–59.1% | 86.9% | 44.4% | 22.0%answered 68% | 17.5written estimate | $2.965$2,965at list prices | 417.016 per request4 in flight |
| 12 | d1‡ Closed · EfficientLatency 0.305 sCost $0.007$7.43 per 1,000per 1M d1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-18) Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.8% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 12.5%, 1 of them wrong Confidence: calibration gap 10.2 points (option score) Cost $0.007$7.43 per 1,000 questionsper 1M questions (free tier, at the list price) · 1 question per request Latency: 0.305 s median per request, 1 in flight | 57.0%54.9%–59.1% | 82.8% | 46.0% | Not run | 10.2option score | $0.007$7.43free tier, at the list price | 0.3051 per request1 in flighton its repeats |
| 12 | Jeffan-9B-1211‡ Open · 9.0BLatency 0.064 sCost $0.044$44.29 per 1,000per 1M Jeffan-9B-1211 · tattrongvu · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.9% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.1% · error rate 42.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.6 points (option score) Cost $0.044$44.29 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.064 s median per request, 1 in flight | 57.0%54.9%–59.1% | 82.9% | 46.0% | 4.3%answered 98% | 4.6option score | $0.044$44.29from GPU time | 0.0641 per request1 in flight |
| 12 | Drex 2.0‡ Open · 35B MoE · 3.0B activeLatency 0.325 sCost $0.019$19.15 per 1,000per 1M Drex 2.0 · Nace.AI · decision model Mixture of experts: 35B total (counted from the released files: 35.11B), about 3B active per step. Size class uses the total (35B). Accuracy 55.7% (95%: 53.6–57.8%) Reading 81.4% · grammar 44.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 4.2 points (option score) Cost $0.019$19.15 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.325 s median per request, 1 in flight | 55.7%53.6%–57.8% | 81.4% | 44.8% | Not run | 4.2option score | $0.019$19.15at list prices | 0.3251 per request1 in flight |
| 12 | Open-Jev-27B-v1.1‡ Open · 27BLatency 0.386 sCost $0.271$271 per 1,000per 1M Open-Jev-27B-v1.1 · Zefan Cai · decision model Base model: Qwen3.8-27B + adapter Parameters: 27B dense, as the maker states it (counted from the released files: 27.80B). Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.9% · grammar 44.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.4% · error rate 44.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 24.5 points (option score) Cost $0.271$271 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.386 s median per request, 1 in flight | 55.7%53.6%–57.8% | 82.9% | 44.1% | 1.4%answered 99% | 24.5option score | $0.271$271from GPU time | 0.3861 per request1 in flighton its repeats |
| 12 | Cygnet‡ Open · 12BLatency 0.036 sCost $0.039$39.39 per 1,000per 1M Cygnet · Blockbrain Labs · decision model Base model: Gemma 4 12B IT Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.8% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 98.5% · error rate 44.2% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 13.8 points (option score) Cost $0.039$39.39 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.036 s median per request, 1 in flight | 55.7%53.6%–57.8% | 82.8% | 44.2% | 3.2%answered 99% | 13.8option score | $0.039$39.39at a host's list price for the same weights | 0.0361 per request1 in flight |
| 12 | Winnow-12B‡ Open · 12BLatency 0.059 sCost $0.041$41.40 per 1,000per 1M Winnow-12B · EldanRing · decision model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.91B). Accuracy 55.5% (95%: 53.4–57.5%) Reading 82.0% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.0% · error rate 44.8% Kept at 99% confidence: 7.4%, 1 of them wrong Confidence: calibration gap 19.7 points (option score) Cost $0.041$41.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 55.5%53.4%–57.5% | 82.0% | 44.2% | 4.3%answered 98% | 19.7option score | $0.041$41.40from GPU time | 0.0591 per request1 in flight |
| 13 | KnowLine-9B-Gen1‡ Open · 9.0BLatency 0.057 sCost $0.041$40.69 per 1,000per 1M KnowLine-9B-Gen1 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy 54.5% (95%: 52.5–56.6%) Reading 79.9% · grammar 43.8% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.8% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 2.8% Confidence: calibration gap 4.6 points (option score) Cost $0.041$40.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.057 s median per request, 1 in flight | 54.5%52.5%–56.6% | 79.9% | 43.8% | 0.5%answered 100% | 4.6option score | $0.041$40.69from GPU time | 0.0571 per request1 in flight |
| 13 | Cerno 9B v1.2‡ Open · 9.0BLatency 0.051 sCost $0.036$36.02 per 1,000per 1M Cerno 9B v1.2 · Apertix AI · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Accuracy 54.5% (95%: 52.4–56.5%) Reading 79.7% · grammar 43.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.2% · answered 99.9% · error rate 45.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.8 points (option score) Cost $0.036$36.02 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.051 s median per request, 1 in flight | 54.5%52.4%–56.5% | 79.7% | 43.7% | 0.2%answered 100% | 9.8option score | $0.036$36.02from GPU time | 0.0511 per request1 in flight |
| 13 | KnowLine-9B-Gen2‡ Open · 9.0BLatency 0.048 sCost $0.035$35.40 per 1,000per 1M KnowLine-9B-Gen2 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy 53.3% (95%: 51.2–55.4%) Reading 79.6% · grammar 42.1% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 2.6% · answered 98.8% · error rate 47.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.1 points (option score) Cost $0.035$35.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.048 s median per request, 1 in flight | 53.3%51.2%–55.4% | 79.6% | 42.1% | 2.6%answered 99% | 5.1option score | $0.035$35.40from GPU time | 0.0481 per request1 in flight |
| 13 | Adjudex 9B v5‡ Open · 9.0BLatency 0.038 sCost $0.026$26.30 per 1,000per 1M Adjudex 9B v5 · newsbang · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Accuracy 53.1% (95%: 51.0–55.2%) Reading 80.8% · grammar 41.4% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.6% · error rate 46.6% Kept at 99% confidence: 6.0%, 2 of them wrong Confidence: calibration gap 9.2 points (option score) Cost $0.026$26.30 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.038 s median per request, 1 in flight | 53.1%51.0%–55.2% | 80.8% | 41.4% | 0.8%answered 100% | 9.2option score | $0.026$26.30from GPU time | 0.0381 per request1 in flight |
| 13 | Elephant-1‡ Open · 9.0BLatency 0.074 sCost $0.014$13.87 per 1,000per 1M Elephant-1 · Base Labs · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.41B). Accuracy 52.4% (95%: 50.3–54.4%) Reading 80.5% · grammar 40.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 34.8% · answered 81.6% · error rate 42.3% Kept at 99% confidence with no observed mistakes: 7.1% Confidence: calibration gap 11.8 points (option score) Cost $0.014$13.87 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.074 s median per request, 1 in flight | 52.4%50.3%–54.4% | 80.5% | 40.4% | 34.8%answered 82% | 11.8option score | $0.014$13.87at the maker's list price | 0.0741 per request1 in flight |
| 13 | GLiDE 9B‡ Open · 7.9BLatency 0.207 sCost No list price GLiDE 9B · Fastino · decision model Parameters: 7.94B dense, as the maker states it (counted from the released files: 7.94B). Accuracy 52.0% (95%: 50.0–54.1%) Reading 80.2% · grammar 40.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 24.1% · answered 87.4% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 8.5 points (option score) Cost: No list price · 1 question per request Latency: 0.207 s median per request, 1 in flight | 52.0%50.0%–54.1% | 80.2% | 40.1% | 24.1%answered 87% | 8.5option score | No list price | 0.2071 per request1 in flight |
| 13 | Dexy-Decision-9B-V2‡ Open · 9.0BLatency 0.031 sCost $0.023$23.03 per 1,000per 1M Dexy-Decision-9B-V2 · Patrick-18 · decision model Base model: vllm-sr/d3-flash Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Accuracy 51.6% (95%: 49.5–53.7%) Reading 79.6% · grammar 39.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.4% · answered 98.3% · error rate 48.4% Kept at 99% confidence: 6.3%, 1 of them wrong Confidence: calibration gap 9.1 points (option score) Cost $0.023$23.03 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.031 s median per request, 1 in flight | 51.6%49.5%–53.7% | 79.6% | 39.7% | 3.4%answered 98% | 9.1option score | $0.023$23.03from GPU time | 0.0311 per request1 in flight |
| 13 | StartLux-Decision-v1.1-9B-Instant‡ Open · 9.0BLatency 0.154 sCost No list price StartLux-Decision-v1.1-9B-Instant · StartLux · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy 51.2% (95%: 49.1–53.3%) Reading 78.8% · grammar 39.5% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 2.2% · answered 98.9% · error rate 49.2% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 7.6 points (option score) Cost: No list price · 1 question per request Latency: 0.154 s median per request, 1 in flight | 51.2%49.1%–53.3% | 78.8% | 39.5% | 2.2%answered 99% | 7.6option score | No list price | 0.1541 per request1 in flight |
| 14 | Perplexity Decider Flash v1.3‡ Open · 4.0BLatency 0.033 sCost $0.025$24.69 per 1,000per 1M Perplexity Decider Flash v1.3 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Accuracy 49.7% (95%: 47.6–51.8%) Reading 74.4% · grammar 39.2% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 6.5% · answered 96.6% · error rate 50.4% Kept at 99% confidence with no observed mistakes: 1.3% Confidence: calibration gap 12.6 points (option score) Cost $0.025$24.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.033 s median per request, 1 in flight | 49.7%47.6%–51.8% | 74.4% | 39.2% | 6.5%answered 97% | 12.6option score | $0.025$24.69from GPU time | 0.0331 per request1 in flight |
| 14 | vera-medium-omni‡ Open · 4.2BLatency 0.029 sCost $0.021$20.72 per 1,000per 1M vera-medium-omni · scar-ai · decision model Base model: Qwen3.5-4B Parameters: 4.2B dense, as the maker states it (counted from the released files: 4.21B). Accuracy 49.1% (95%: 47.0–51.2%) Reading 75.5% · grammar 37.9% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 8.9% · answered 95.2% · error rate 51.4% Kept at 99% confidence with no observed mistakes: 0.2% Confidence: calibration gap 3.7 points (option score) Cost $0.021$20.72 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.029 s median per request, 1 in flight | 49.1%47.0%–51.2% | 75.5% | 37.9% | 8.9%answered 95% | 3.7option score | $0.021$20.72from GPU time | 0.0291 per request1 in flight |
| 14 | KnowLine 4B Gen5.5‡ Open · 4.0BLatency 0.099 sCost No list price KnowLine 4B Gen5.5 · PelaAI · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy 48.5% (95%: 46.4–50.6%) Reading 74.8% · grammar 37.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 97.5% · error rate 52.3% Kept at 99% confidence with no observed mistakes: 1.8% Confidence: calibration gap 13.5 points (option score) Cost: No list price · 1 question per request Latency: 0.099 s median per request, 1 in flight | 48.5%46.4%–50.6% | 74.8% | 37.3% | 4.7%answered 97% | 13.5option score | No list price | 0.0991 per request1 in flight |
| 14 | StartLux-Decision-v1.1-4B-Instant‡ Open · 4.0BLatency 0.074 sCost No list price StartLux-Decision-v1.1-4B-Instant · StartLux · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy 47.9% (95%: 45.8–50.0%) Reading 74.1% · grammar 36.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.4% Confidence: calibration gap 7.1 points (option score) Cost: No list price · 1 question per request Latency: 0.074 s median per request, 1 in flight | 47.9%45.8%–50.0% | 74.1% | 36.7% | 1.8%answered 99% | 7.1option score | No list price | 0.0741 per request1 in flight |
| 14 | Perplexity Decider Flash v1.2‡ Open · 4.0BLatency 0.132 sCost $0.004$3.88 per 1,000per 1M Perplexity Decider Flash v1.2 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Accuracy 47.8% (95%: 45.7–49.9%) Reading 72.7% · grammar 37.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.7% · answered 91.8% · error rate 51.7% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost $0.004$3.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.132 s median per request, 1 in flight | 47.8%45.7%–49.9% | 72.7% | 37.2% | 14.7%answered 92% | 8.3option score | $0.004$3.88at the maker's list price | 0.1321 per request1 in flight |
| 14 | Valen-Decision-1.5-9B‡ Open · 9.0BLatency 0.059 sCost $0.040$40.49 per 1,000per 1M Valen-Decision-1.5-9B · Valen-Team · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.40B). Accuracy 47.0% (95%: 44.9–49.0%) Reading 75.0% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.3% Confidence: calibration gap 3.8 points (option score) Cost $0.040$40.49 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 47.0%44.9%–49.0% | 75.0% | 35.0% | 0.1%answered 100% | 3.8option score | $0.040$40.49from GPU time | 0.0591 per request1 in flight |
| 14 | Verdin-4B-v1.1‡ Open · 4.0BLatency 0.049 sCost $0.034$34.21 per 1,000per 1M Verdin-4B-v1.1 · victorbennett · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Accuracy 46.8% (95%: 44.7–48.9%) Reading 71.2% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.8% · error rate 53.3% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 9.8 points (option score) Cost $0.034$34.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.049 s median per request, 1 in flight | 46.8%44.7%–48.9% | 71.2% | 36.4% | 0.4%answered 100% | 9.8option score | $0.034$34.21from GPU time | 0.0491 per request1 in flight |
| 14 | Chinese Decision Pointer 4B‡ Open · 4.0BLatency 0.035 sCost $0.025$24.58 per 1,000per 1M Chinese Decision Pointer 4B · quincyqiang · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Accuracy 46.7% (95%: 44.6–48.8%) Reading 70.9% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 11.2% · answered 93.8% · error rate 52.5% Kept at 99% confidence: 2.3%, 1 of them wrong Confidence: calibration gap 11.7 points (option score) Cost $0.025$24.58 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.035 s median per request, 1 in flight | 46.7%44.6%–48.8% | 70.9% | 36.4% | 11.2%answered 94% | 11.7option score | $0.025$24.58from GPU time | 0.0351 per request1 in flight |
| 14 | Microsoft-Decision-1‡ Closed · EfficientLatency 0.286 sCost $0.014$13.82 per 1,000per 1M Microsoft-Decision-1 · Microsoft · decision model Base model: Qwen3.5-9B Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://openrouter.ai/microsoft/microsoft-decision-1, read 2026-10-17) Accuracy 46.5% (95%: 44.4–48.5%) Reading 73.5% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 2.8%, 3 of them wrong Confidence: calibration gap 19.8 points (option score) Cost $0.014$13.82 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.286 s median per request, 1 in flight | 46.5%44.4%–48.5% | 73.5% | 35.0% | Not run | 19.8option score | $0.014$13.82at list prices | 0.2861 per request1 in flight |
| 14 | Qwen3.5-4B reasoning off · chat‡ Open · 4.0BLatency 32.2 sCost $0.894$894 per 1,000per 1M Qwen3.5-4B reasoning off · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy 46.2% (95%: 44.1–48.3%) Reading 68.4% · grammar 36.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.9% · answered 95.4% · error rate 54.0% Confidence: not measured Cost $0.894$894 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 32.2 s median per request, 32 in flight | 46.2%44.1%–48.3% | 68.4% | 36.7% | 6.9%answered 95% | Not measureda plain answer line, no confidence | $0.894$894from GPU time | 32.21 per request32 in flight |
| 14 | Qwen3-14B reasoning off · chat‡ Open · 15BLatency 2.85 sCost $0.032$32.31 per 1,000per 1M Qwen3-14B reasoning off · chat · Alibaba · chat model Parameters: 14.8B dense, as the maker states it (counted from the released files: 14.77B). Accuracy 45.7% (95%: 43.6–47.8%) Reading 72.1% · grammar 34.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.0% · error rate 53.8% Confidence: calibration gap 41.0 points (answer letter probability) Cost $0.032$32.31 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 2.85 s median per request, 1 in flight | 45.7%43.6%–47.8% | 72.1% | 34.5% | 1.9%answered 99% | 41.0answer letter probability | $0.032$32.31at a host's list price for the same weights | 2.8516 per request1 in flight |
| 14 | Clef-Flash‡ Open · 9.0BLatency 0.075 sCost $0.017$17.37 per 1,000per 1M Clef-Flash · Cloudflare · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.53B). Accuracy 45.7% (95%: 43.6–47.8%) Reading 73.0% · grammar 34.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.0% · answered 95.0% · error rate 53.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 15.5 points (option score) Cost $0.017$17.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.075 s median per request, 1 in flight | 45.7%43.6%–47.8% | 73.0% | 34.1% | 9.0%answered 95% | 15.5option score | $0.017$17.37at the maker's list price | 0.0751 per request1 in flight |
| 14 | Erk-14B reasoning off · chat Open · 14BLatency 2.39 sCost $0.106$106 per 1,000per 1M Erk-14B reasoning off · chat · eCloud Technology · chat model Parameters: 14B dense, as the maker states it (counted from the released files: 14.77B). Accuracy 45.0% (95%: 42.9–47.1%) Reading 72.3% · grammar 33.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.8% · error rate 55.3% Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.106$106 per 1,000 questionsper 1M questions (from GPU time) · 16 questions per request Latency: 2.39 s median per request, 1 in flight | 45.0%42.9%–47.1% | 72.3% | 33.4% | 0.3%answered 100% | 13.5answer letter probability | $0.106$106from GPU time | 2.3916 per request1 in flight |
| 15 | Metask-Jev-4B Open · 4.5BLatency 0.090 sCost $0.063$62.82 per 1,000per 1M Metask-Jev-4B · Metask Lab · decision model Parameters: 4.54B dense, as the maker states it (counted from the released files: 4.54B). Accuracy 40.8% (95%: 38.7–42.8%) Reading 62.8% · grammar 31.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.5% · error rate 60.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.063$62.82 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.090 s median per request, 1 in flight | 40.8%38.7%–42.8% | 62.8% | 31.4% | 0.8%answered 100% | 5.8option score | $0.063$62.82from GPU time | 0.0901 per request1 in flight |
| 15 | KnowLine-2B-Gen1‡ Open · 2.0BLatency 0.062 sCost No list price KnowLine-2B-Gen1 · PelaAI · decision model Base model: Qwen3.5-2B Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy 36.8% (95%: 34.8–38.8%) Reading 54.3% · grammar 29.4% ‡ Turkish not officially supported: it lists other languages (source, read 13 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.4 points (option score) Cost: No list price · 1 question per request Latency: 0.062 s median per request, 1 in flight | 36.8%34.8%–38.8% | 54.3% | 29.4% | Not run | 8.4option score | No list price | 0.0621 per request1 in flight |
| 16 | GLiDE 2B‡ Open · 1.9BLatency 0.034 sCost No list price GLiDE 2B · Fastino · decision model Parameters: 1.88B dense, as the maker states it (counted from the released files: 1.88B). Accuracy 35.3% (95%: 33.3–37.3%) Reading 51.2% · grammar 28.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.5% · answered 72.2% · error rate 61.7% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 20.7 points (option score) Cost: No list price · 1 question per request Latency: 0.034 s median per request, 1 in flight | 35.3%33.3%–37.3% | 51.2% | 28.5% | 38.5%answered 72% | 20.7option score | No list price | 0.0341 per request1 in flight |
| 16 | StartLux-Decision-v1.1-2B-Instant‡ Open · 2.0BLatency 0.040 sCost No list price StartLux-Decision-v1.1-2B-Instant · StartLux · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy 34.3% (95%: 32.3–36.3%) Reading 52.9% · grammar 26.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 21.5% · answered 84.8% · error rate 65.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.4 points (option score) Cost: No list price · 1 question per request Latency: 0.040 s median per request, 1 in flight | 34.3%32.3%–36.3% | 52.9% | 26.3% | 21.5%answered 85% | 6.4option score | No list price | 0.0401 per request1 in flight |
| 16 | LFM2.5-2.6B default reasoning · chat‡ Open · 2.7BLatency 51.2 sCost $2.761$2,761 per 1,000per 1M LFM2.5-2.6B default reasoning · chat · Liquid AI · chat model Parameters: 2.69B dense, as the maker states it (counted from the released files: 2.70B). Accuracy 33.9% (95%: 31.9–35.9%) Reading 45.4% · grammar 29.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 24.7% · answered 81.6% · error rate 63.2% Confidence: not measured Cost $2.761$2,761 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 51.2 s median per request, 32 in flight | 33.9%31.9%–35.9% | 45.4% | 29.0% | 24.7%answered 82% | Not measureda plain answer line, no confidence | $2.761$2,761at the maker's list price | 51.21 per request32 in flight |
| 16 | LFM3-1.7B reasoning on · chat‡ Open · 1.7BLatency 24.2 sCost $1.277$1,277 per 1,000per 1M LFM3-1.7B reasoning on · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Accuracy 32.4% (95%: 30.5–34.4%) Reading 44.2% · grammar 27.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 67.0% · answered 45.5% · error rate 58.8% Confidence: not measured Cost $1.277$1,277 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 24.2 s median per request, 32 in flight | 32.4%30.5%–34.4% | 44.2% | 27.4% | 67.0%answered 46% | Not measureda plain answer line, no confidence | $1.277$1,277at the maker's list price | 24.21 per request32 in flight |
| 17 | d1-2B‡ Open · 1.9BLatency 0.019 sCost $0.013$13.17 per 1,000per 1M d1-2B · Liquid AI · decision model Base model: LFM3-VL-2B Parameters: 1.92B dense, as the maker states it (counted from the released files: 1.92B). Accuracy 29.4% (95%: 27.5–31.3%) Reading 40.9% · grammar 24.5% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 65.5% · answered 45.0% · error rate 64.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.013$13.17 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.019 s median per request, 1 in flight | 29.4%27.5%–31.3% | 40.9% | 24.5% | 65.5%answered 45% | 5.8option score | $0.013$13.17from GPU time | 0.0191 per request1 in flight |
| 17 | GLiDE 0.8B‡ Open · 0.8BLatency 0.022 sCost No list price GLiDE 0.8B · Fastino · decision model Parameters: 750M dense, as the maker states it (counted from the released files: 0.75B). Accuracy 29.3% (95%: 27.4–31.2%) Reading 42.4% · grammar 23.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.2% · answered 78.2% · error rate 67.7% Kept at 99% confidence: 0.8%, 1 of them wrong Confidence: calibration gap 24.8 points (option score) Cost: No list price · 1 question per request Latency: 0.022 s median per request, 1 in flight | 29.3%27.4%–31.2% | 42.4% | 23.7% | 29.2%answered 78% | 24.8option score | No list price | 0.0221 per request1 in flight |
| 17 | OneJev1.6-0.8B‡ Open · 0.8BLatency 0.017 sCost No list price OneJev1.6-0.8B · OmniJev · decision model Parameters: 800M dense, as the maker states it (counted from the released files: 0.85B). Accuracy 29.1% (95%: 27.3–31.1%) Reading 39.6% · grammar 24.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 43.4% · answered 65.7% · error rate 68.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.3 points (option score) Cost: No list price · 1 question per request Latency: 0.017 s median per request, 1 in flight | 29.1%27.3%–31.1% | 39.6% | 24.6% | 43.4%answered 66% | 9.3option score | No list price | 0.0171 per request1 in flight |
| 17 | StartLux-Decision-v1.1-0.8B-Instant‡ Open · 0.8BLatency 0.028 sCost No list price StartLux-Decision-v1.1-0.8B-Instant · StartLux · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Accuracy 28.6% (95%: 26.7–30.5%) Reading 35.1% · grammar 25.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 12.4% · answered 90.9% · error rate 70.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.4 points (option score) Cost: No list price · 1 question per request Latency: 0.028 s median per request, 1 in flight | 28.6%26.7%–30.5% | 35.1% | 25.8% | 12.4%answered 91% | 4.4option score | No list price | 0.0281 per request1 in flight |
| 17 | LFM3-1.7B reasoning off · chat‡ Open · 1.7BLatency 4.90 sCost $0.646$646 per 1,000per 1M LFM3-1.7B reasoning off · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Accuracy 28.3% (95%: 26.5–30.3%) Reading 37.2% · grammar 24.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 76.2% · answered 32.7% · error rate 62.4% Confidence: not measured Cost $0.646$646 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 4.90 s median per request, 32 in flight | 28.3%26.5%–30.3% | 37.2% | 24.6% | 76.2%answered 33% | Not measureda plain answer line, no confidence | $0.646$646at the maker's list price | 4.901 per request32 in flight |
| 17 | Strands Decider 2B‡ Open · 2.0BLatency 0.026 sCost $0.020$20.12 per 1,000per 1M Strands Decider 2B · Strands Agents · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.29B). Accuracy 27.5% (95%: 25.7–29.4%) Reading 36.9% · grammar 23.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 93.2% · error rate 72.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 12.2 points (option score) Cost $0.020$20.12 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.026 s median per request, 1 in flight | 27.5%25.7%–29.4% | 36.9% | 23.5% | 9.2%answered 93% | 12.2option score | $0.020$20.12from GPU time | 0.0261 per request1 in flight |
| 17 | Decider 1‡ Closed · EfficientLatency 0.382 sCost $0.015$14.76 per 1,000per 1M Decider 1 · meraGPT · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://meragpt.com/docs/models, read 2026-10-19) Accuracy 27.1% (95%: 25.3–29.0%) Reading 34.6% · grammar 23.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 14.3 points (option score) Cost $0.015$14.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.382 s median per request, 1 in flight | 27.1%25.3%–29.0% | 34.6% | 23.9% | Not run | 14.3option score | $0.015$14.76at list prices | 0.3821 per request1 in flight |
| 17 | KnowLine-0.8B-Gen1‡ Open · 0.8BLatency 0.033 sCost No list price KnowLine-0.8B-Gen1 · PelaAI · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Accuracy 27.1% (95%: 25.3–29.0%) Reading 38.3% · grammar 22.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 95.3% · error rate 72.3% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost: No list price · 1 question per request Latency: 0.033 s median per request, 1 in flight | 27.1%25.3%–29.0% | 38.3% | 22.4% | 6.4%answered 95% | 8.3option score | No list price | 0.0331 per request1 in flight |
| 17 | Qwen3.5-2B reasoning off · chat‡ Open · 2.0BLatency 8.84 sCost $0.268$268 per 1,000per 1M Qwen3.5-2B reasoning off · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy 25.4% (95%: 23.6–27.2%) Reading 34.3% · grammar 21.6% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 20.8% · answered 75.6% · error rate 72.3% Confidence: not measured Cost $0.268$268 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 8.84 s median per request, 32 in flight | 25.4%23.6%–27.2% | 34.3% | 21.6% | 20.8%answered 76% | Not measureda plain answer line, no confidence | $0.268$268from GPU time | 8.841 per request32 in flight |
| 18 | GLiClass v4 Edge‡ Open · 0.4BLatency 0.025 sCost No list price GLiClass v4 Edge · Knowledgator · decision model Base model: DeBERTa-v3-large Parameters: 400M dense, as the maker states it (counted from the released files: 0.44B). Accuracy 23.3% (95%: 21.6–25.2%) Reading 25.9% · grammar 22.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.3% · answered 81.9% · error rate 77.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.5 points (option score) Cost: No list price · 1 question per request Latency: 0.025 s median per request, 1 in flight | 23.3%21.6%–25.2% | 25.9% | 22.2% | 22.3%answered 82% | 10.5option score | No list price | 0.0251 per request1 in flight |
| 18 | GLiDE 0.4B‡ Open · 0.4BLatency 0.016 sCost No list price GLiDE 0.4B · Fastino · decision model Parameters: 400M dense, as the maker states it (counted from the released files: 0.40B). Accuracy 22.2% (95%: 20.6–24.0%) Reading 24.5% · grammar 21.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 95.3% · answered 5.8% · error rate 78.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.0 points (option score) Cost: No list price · 1 question per request Latency: 0.016 s median per request, 1 in flight | 22.2%20.6%–24.0% | 24.5% | 21.3% | 95.3%answered 6% | 8.0option score | No list price | 0.0161 per request1 in flight |
| 18 | LFM2.5-8B-A1B default reasoning · chat‡ Open · 8.3B MoE · 1.5B activeLatency 39.2 sCost $1.050$1,050 per 1,000per 1M LFM2.5-8B-A1B default reasoning · chat · Liquid AI · chat model Mixture of experts: 8.3B total (counted from the released files: 8.47B), about 1.5B active per step. Size class uses the total (8.3B). Accuracy 21.0% (95%: 19.3–22.7%) Reading 25.6% · grammar 19.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 56.0% · answered 42.4% · error rate 74.0% Confidence: not measured Cost $1.050$1,050 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 39.2 s median per request, 32 in flight | 21.0%19.3%–22.7% | 25.6% | 19.0% | 56.0%answered 42% | Not measureda plain answer line, no confidence | $1.050$1,050from GPU time | 39.21 per request32 in flight |
| 18 | GLiNER2.5-multi-Decide‡ Open · 0.3BLatency 0.009 sCost $0.006$6.27 per 1,000per 1M GLiNER2.5-multi-Decide · Fastino · decision model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Accuracy 20.2% (95%: 18.6–21.9%) Reading 21.0% · grammar 19.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 94.8% · error rate 79.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 23.1 points (option score) Cost $0.006$6.27 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 20.2%18.6%–21.9% | 21.0% | 19.8% | 6.4%answered 95% | 23.1option score | $0.006$6.27from GPU time | 0.0091 per request1 in flight |
| 18 | Laya‡ Open · 0.4BLatency 0.011 sCost $0.009$9.00 per 1,000per 1M Laya · Convai Innovations · decision model Parameters: 421M dense, as the maker states it (counted from the released files: 0.42B). Accuracy 19.0% (95%: 17.4–20.7%) Reading 20.9% · grammar 18.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 28.3% · answered 75.6% · error rate 81.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 13.2 points (option score) Cost $0.009$9.00 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.011 s median per request, 1 in flight | 19.0%17.4%–20.7% | 20.9% | 18.2% | 28.3%answered 76% | 13.2option score | $0.009$9.00from GPU time | 0.0111 per request1 in flighton its repeats |
| 19 | Laya Multilingual‡ Open · 0.3BLatency 0.009 sCost $0.008$7.74 per 1,000per 1M Laya Multilingual · Convai Innovations · decision model Parameters: 322M dense, as the maker states it (counted from the released files: 0.32B). Accuracy 17.9% (95%: 16.3–19.5%) Reading 17.1% · grammar 18.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.8% · answered 87.3% · error rate 83.8% Kept at 99% confidence: 0.0%, 1 of them wrong Confidence: calibration gap 27.7 points (option score) Cost $0.008$7.74 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 17.9%16.3%–19.5% | 17.1% | 18.2% | 14.8%answered 87% | 27.7option score | $0.008$7.74from GPU time | 0.0091 per request1 in flighton its repeats |
| 19 | GLiNER2.5 Multi‡ Open · 0.3BLatency 0.032 sCost $0.006$5.88 per 1,000per 1M GLiNER2.5 Multi · Fastino · chat model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Accuracy 17.5% (95%: 15.9–19.1%) Reading 17.2% · grammar 17.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.3% · answered 94.7% · error rate 82.6% Confidence: calibration gap 35.6 points (option score) Cost $0.006$5.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.032 s median per request, 1 in flight | 17.5%15.9%–19.1% | 17.2% | 17.6% | 6.3%answered 95% | 35.6option score | $0.006$5.88at the maker's list price | 0.0321 per request1 in flight |
| 20 | MiniCPM5-2B reasoning off · chat‡ Open · 2.5BLatency 172.0 sCost $5.464$5,464 per 1,000per 1M MiniCPM5-2B reasoning off · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Accuracy 5.6% (95%: 4.7–6.6%) Reading 6.6% · grammar 5.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 25.4% · answered 20.1% · error rate 75.3% Confidence: not measured Cost $5.464$5,464 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 172.0 s median per request, 32 in flight | 5.6%4.7%–6.6% | 6.6% | 5.2% | 25.4%answered 20% | Not measureda plain answer line, no confidence | $5.464$5,464from GPU time | 172.01 per request32 in flight |
Accuracy: the share of the 2,198 Turkish questions answered right, with its 95% interval as the whisker; higher is better.
No row of this part matches these keys. The small open chat models ran in English only.
Tier: ranked by accuracy among the rows in view, a row joins the tier of the tier's top row when the all-pairs test behind the Accuracy section's letters (9,316 pairs, exact McNemar, Holm-corrected, 5%) found no difference between them, else it starts the next tier; a line separates tiers. Rows in one tier need not be equal. A setup with web search is in no tier (Tier: –).
‡ Turkish not officially supported: the model's own card or maker docs do not list Turkish (read 16 November 2026; Ümit #5067); of the models here only Erk-14B and Metask-Jev-4B list it.
| Rank | Model | Accuracy on readingof 656 questions, 95% interval | Turkishof 2,198 questions | Grammarof 1,542 questions | Non-hallucination% “I don't know” of answers not right | Confidencecalibration gap, points | Costper 1,000 questionsper 1M questions | Latencymedian s per request |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 max · chat‡ Closed · FlagshipLatency 81.1 sCost $53.381$53,381 per 1,000per 1M Claude Fable 5.2 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $8.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-11-11) Turkish reading 97.7% (95%: 96.3–98.6%) Accuracy 98.1% (95%: 97.5–98.6%) Reading 97.7% · grammar 98.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know”: Queued (its run with the option has not started yet) Confidence: calibration gap 7.5 points (written estimate) Cost $53.381$53,381 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 81.1 s median per request, 1 in flight | 97.7%96.3%–98.6% | 98.1% | 98.3% | Queued | 7.5written estimate | $53.381$53,381subscription, priced as the API | 81.116 per request1 in flight |
| 2 | Gemini 4 Carbon high · chat‡ Closed · FlagshipLatency 217.7 sCost $31.309$31,309 per 1,000per 1M Gemini 4 Carbon high · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish reading 97.6% (95%: 96.1–98.5%) Accuracy 98.5% (95%: 97.8–98.9%) Reading 97.6% · grammar 98.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.9% · answered 99.7% · error rate 1.5% Confidence: calibration gap 0.4 points (written estimate) Cost $31.309$31,309 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 217.7 s median per request, 1 in flight | 97.6%96.1%–98.5% | 98.5% | 98.8% | 17.9%answered 100% | 0.4written estimate | $31.309$31,309subscription, priced as the API | 217.716 per request1 in flight |
| 3 | Gemini 4 Hydrogen high · chat‡ Closed · WorkhorseLatency 31.7 sCost $4.806$4,806 per 1,000per 1M Gemini 4 Hydrogen high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish reading 97.3% (95%: 95.7–98.3%) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.3% · grammar 98.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.4% · answered 99.1% · error rate 1.7% Confidence: calibration gap 1.2 points (written estimate) Cost $4.806$4,806 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 31.7 s median per request, 1 in flight | 97.3%95.7%–98.3% | 98.1% | 98.4% | 17.4%answered 99% | 1.2written estimate | $4.806$4,806subscription, priced as the API | 31.716 per request1 in flight |
| 4 | Gemini 3.8 Flash high · chat‡ Closed · WorkhorseLatency 53.7 sCost $2.419$2,419 per 1,000per 1M Gemini 3.8 Flash high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish reading 97.1% (95%: 95.5–98.1%) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.1% · grammar 98.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.5% · answered 99.9% · error rate 1.9% Confidence: calibration gap 1.1 points (written estimate) Cost $2.419$2,419 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 53.7 s median per request, 4 in flight | 97.1%95.5%–98.1% | 98.1% | 98.5% | 4.5%answered 100% | 1.1written estimate | $2.419$2,419at list prices | 53.716 per request4 in flight |
| Gemini 3.8 Flash low · chat‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.19 sCost $0.269$269 per 1,000per 1M Gemini 3.8 Flash low · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish reading 95.1% (95%: 93.2–96.5%) Accuracy 93.3% (95%: 92.1–94.2%) Reading 95.1% · grammar 92.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.9% · error rate 7.2% Confidence: calibration gap 3.5 points (written estimate) Cost $0.269$269 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 2.19 s median per request, 1 in flight | 95.1%93.2%–96.5% | 93.3% | 92.5% | 1.9%answered 100% | 3.5written estimate | $0.269$269at list prices | 2.1916 per request1 in flight | |
| Gemini 3.8 Flash low · chat · 1 per request‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.72 sCost $0.963$963 per 1,000per 1M Gemini 3.8 Flash low · chat · 1 per request · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish reading 96.2% (95%: 94.4–97.4%) Accuracy 97.1% (95%: 96.3–97.8%) Reading 96.2% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.2% · answered 99.8% · error rate 2.7% Confidence: calibration gap 0.6 points (written estimate) Cost $0.963$963 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.72 s median per request, 1 in flight | 96.2%94.4%–97.4% | 97.1% | 97.5% | 6.2%answered 100% | 0.6written estimate | $0.963$963at list prices | 2.721 per request1 in flight | |
| 5 | GPT-6 Astra high · chat‡ Closed · FlagshipLatency 21.7 sCost $6.066$6,066 per 1,000per 1M GPT-6 Astra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 97.1% (95%: 95.5–98.1%) Accuracy 98.0% (95%: 97.3–98.5%) Reading 97.1% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.3% · answered 99.7% · error rate 1.9% Confidence: calibration gap 0.7 points (written estimate) Cost $6.066$6,066 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 21.7 s median per request, 1 in flight | 97.1%95.5%–98.1% | 98.0% | 98.3% | 14.3%answered 100% | 0.7written estimate | $6.066$6,066subscription, priced as the API | 21.716 per request1 in flight |
| GPT-6 Astra medium · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 15.3 sCost $5.447$5,447 per 1,000per 1M GPT-6 Astra medium · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 97.0% (95%: 95.3–98.0%) Accuracy 97.5% (95%: 96.8–98.1%) Reading 97.0% · grammar 97.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 12.7% · answered 99.7% · error rate 2.2% Confidence: calibration gap 1.0 points (written estimate) Cost $5.447$5,447 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 15.3 s median per request, 1 in flight | 97.0%95.3%–98.0% | 97.5% | 97.7% | 12.7%answered 100% | 1.0written estimate | $5.447$5,447subscription, priced as the API | 15.316 per request1 in flight | |
| GPT-6 Astra low · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 13.0 sCost $5.378$5,378 per 1,000per 1M GPT-6 Astra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 97.0% (95%: 95.3–98.0%) Accuracy 97.6% (95%: 96.9–98.2%) Reading 97.0% · grammar 97.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 10.5% · answered 99.7% · error rate 2.3% Confidence: calibration gap 0.8 points (written estimate) Cost $5.378$5,378 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.0 s median per request, 1 in flight | 97.0%95.3%–98.0% | 97.6% | 97.9% | 10.5%answered 100% | 0.8written estimate | $5.378$5,378subscription, priced as the API | 13.016 per request1 in flighton a sample | |
| GPT-6 Astra low · chat · 1 per request‡ Closed · Flagshipother setting of GPT-6 AstraLatency 6.41 sCost $26.738$26,738 per 1,000per 1M GPT-6 Astra low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 97.3% (95%: 95.7–98.3%) Accuracy 97.4% (95%: 96.7–98.0%) Reading 97.3% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.9% · answered 99.7% · error rate 2.4% Confidence: calibration gap 1.3 points (written estimate) Cost $26.738$26,738 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.41 s median per request, 1 in flight | 97.3%95.7%–98.3% | 97.4% | 97.5% | 11.9%answered 100% | 1.3written estimate | $26.738$26,738subscription, priced as the API | 6.411 per request1 in flight | |
| 6 | Gemini 4 Argon low · chat‡ Closed · FlagshipLatency 5.46 sCost $1.786$1,786 per 1,000per 1M Gemini 4 Argon low · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-28) Turkish reading 97.0% (95%: 95.3–98.0%) Accuracy 96.4% (95%: 95.5–97.1%) Reading 97.0% · grammar 96.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.0% · answered 99.7% · error rate 3.6% Confidence: calibration gap 2.2 points (written estimate) Cost $1.786$1,786 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.46 s median per request, 1 in flight | 97.0%95.3%–98.0% | 96.4% | 96.1% | 7.0%answered 100% | 2.2written estimate | $1.786$1,786at list prices | 5.4616 per request1 in flight |
| 7 | GPT-6.1 Sol high · chat‡ Closed · WorkhorseLatency 24.7 sCost $1.622$1,622 per 1,000per 1M GPT-6.1 Sol high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 96.5% (95%: 94.8–97.7%) Accuracy 97.8% (95%: 97.1–98.3%) Reading 96.5% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 15.4% · answered 99.6% · error rate 2.0% Confidence: calibration gap 1.0 points (written estimate) Cost $1.622$1,622 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 24.7 s median per request, 1 in flight | 96.5%94.8%–97.7% | 97.8% | 98.3% | 15.4%answered 100% | 1.0written estimate | $1.622$1,622subscription, priced as the API | 24.716 per request1 in flight |
| GPT-6.1 Sol low · chat‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 13.3 sCost $1.130$1,130 per 1,000per 1M GPT-6.1 Sol low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 96.2% (95%: 94.4–97.4%) Accuracy 96.7% (95%: 95.8–97.4%) Reading 96.2% · grammar 96.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 99.7% · error rate 2.7% Confidence: calibration gap 1.5 points (written estimate) Cost $1.130$1,130 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.3 s median per request, 1 in flight | 96.2%94.4%–97.4% | 96.7% | 96.9% | 9.2%answered 100% | 1.5written estimate | $1.130$1,130subscription, priced as the API | 13.316 per request1 in flight | |
| GPT-6.1 Sol low · chat · 1 per request‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 6.61 sCost $5.569$5,569 per 1,000per 1M GPT-6.1 Sol low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish reading 96.0% (95%: 94.3–97.3%) Accuracy 96.7% (95%: 95.9–97.4%) Reading 96.0% · grammar 97.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.6% · answered 99.6% · error rate 2.8% Confidence: calibration gap 1.8 points (written estimate) Cost $5.569$5,569 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.61 s median per request, 1 in flight | 96.0%94.3%–97.3% | 96.7% | 97.0% | 11.6%answered 100% | 1.8written estimate | $5.569$5,569subscription, priced as the API | 6.611 per request1 in flight | |
| 8 | GPT-6.1 Terra high · chat‡ Closed · WorkhorseLatency 7.10 sCost $0.515$515 per 1,000per 1M GPT-6.1 Terra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish reading 96.2% (95%: 94.4–97.4%) Accuracy 94.3% (95%: 93.2–95.2%) Reading 96.2% · grammar 93.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.3% · answered 99.6% · error rate 5.3% Confidence: calibration gap 2.8 points (written estimate) Cost $0.515$515 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.10 s median per request, 1 in flight | 96.2%94.4%–97.4% | 94.3% | 93.5% | 7.3%answered 100% | 2.8written estimate | $0.515$515subscription, priced as the API | 7.1016 per request1 in flight |
| GPT-6.1 Terra low · chat‡ Closed · Workhorseother setting of GPT-6.1 TerraLatency 4.24 sCost $0.426$426 per 1,000per 1M GPT-6.1 Terra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish reading 95.9% (95%: 94.1–97.2%) Accuracy 90.9% (95%: 89.6–92.0%) Reading 95.9% · grammar 88.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.9% · answered 99.7% · error rate 9.1% Confidence: calibration gap 6.0 points (written estimate) Cost $0.426$426 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 4.24 s median per request, 1 in flight | 95.9%94.1%–97.2% | 90.9% | 88.7% | 2.9%answered 100% | 6.0written estimate | $0.426$426subscription, priced as the API | 4.2416 per request1 in flight | |
| 9 | Claude Opus 5.5 high · chat‡ Closed · FlagshipLatency 22.1 sCost $6.509$6,509 per 1,000per 1M Claude Opus 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 95.6% (95%: 93.7–96.9%) Accuracy 95.0% (95%: 94.0–95.8%) Reading 95.6% · grammar 94.7% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 2.7% · answered 99.9% · error rate 4.9% Confidence: calibration gap 14.6 points (written estimate) Cost $6.509$6,509 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 22.1 s median per request, 1 in flight | 95.6%93.7%–96.9% | 95.0% | 94.7% | 2.7%answered 100% | 14.6written estimate | $6.509$6,509subscription, priced as the API | 22.116 per request1 in flight |
| Claude Opus 5.5 low · chat‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 13.4 sCost $4.931$4,931 per 1,000per 1M Claude Opus 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 94.8% (95%: 92.8–96.3%) Accuracy 90.4% (95%: 89.1–91.6%) Reading 94.8% · grammar 88.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.9% · error rate 7.7% Confidence: calibration gap 16.5 points (written estimate) Cost $4.931$4,931 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.4 s median per request, 1 in flight | 94.8%92.8%–96.3% | 90.4% | 88.6% | 1.8%answered 100% | 16.5written estimate | $4.931$4,931subscription, priced as the API | 13.416 per request1 in flighton a sample | |
| Claude Opus 5.5 low · chat · 1 per request‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 3.71 sCost $11.610$11,610 per 1,000per 1M Claude Opus 5.5 low · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 95.7% (95%: 93.9–97.0%) Accuracy 93.3% (95%: 92.2–94.3%) Reading 95.7% · grammar 92.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 100.0% · error rate 6.2% Confidence: calibration gap 10.2 points (written estimate) Cost $11.610$11,610 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 3.71 s median per request, 1 in flight | 95.7%93.9%–97.0% | 93.3% | 92.3% | 0.7%answered 100% | 10.2written estimate | $11.610$11,610subscription, priced as the API | 3.711 per request1 in flight | |
| 10 | Claude Fable 5.1 high · chat‡ Closed · FlagshipLatency 30.3 sCost $17.042$17,042 per 1,000per 1M Claude Fable 5.1 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-28) Turkish reading 95.6% (95%: 93.7–96.9%) Accuracy 96.1% (95%: 95.2–96.9%) Reading 95.6% · grammar 96.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.0% · answered 99.8% · error rate 3.6% Confidence: calibration gap 12.0 points (written estimate) Cost $17.042$17,042 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 30.3 s median per request, 1 in flight | 95.6%93.7%–96.9% | 96.1% | 96.4% | 6.0%answered 100% | 12.0written estimate | $17.042$17,042subscription, priced as the API | 30.316 per request1 in flight |
| 11 | Grok 4.7 high · chat‡ Closed · WorkhorseLatency 401.9 sCost $14.292$14,292 per 1,000per 1M Grok 4.7 high · chat · xAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://docs.x.ai/docs/models/grok-4.7, read 2026-10-28) Turkish reading 95.0% (95%: 93.0–96.4%) Accuracy 92.7% (95%: 91.6–93.7%) Reading 95.0% · grammar 91.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.9% · answered 97.5% · error rate 6.1% Confidence: calibration gap 11.9 points (written estimate) Cost $14.292$14,292 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 401.9 s median per request, 4 in flight | 95.0%93.0%–96.4% | 92.7% | 91.8% | 29.9%answered 97% | 11.9written estimate | $14.292$14,292subscription, priced as the API | 401.916 per request4 in flight |
| 12 | Qwen4-Max high · chat‡ Closed · WorkhorseLatency 91.5 sCost $2.533$2,533 per 1,000per 1M Qwen4-Max high · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://openrouter.ai/qwen/qwen4-max, read 2026-11-16) Turkish reading 94.7% (95%: 92.7–96.1%) Accuracy 91.7% (95%: 90.5–92.8%) Reading 94.7% · grammar 90.5% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.2% · error rate 6.4% Confidence: calibration gap 7.8 points (written estimate) Cost $2.533$2,533 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 91.5 s median per request, 4 in flight | 94.7%92.7%–96.1% | 91.7% | 90.5% | 0.7%answered 99% | 7.8written estimate | $2.533$2,533at list prices | 91.516 per request4 in flight |
| 13 | Kimi K3.1 Pro low · chat‡ Open · 2.8T MoE · 104B activeLatency 44.9 sCost $2.048$2,048 per 1,000per 1M Kimi K3.1 Pro low · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish reading 93.8% (95%: 91.6–95.4%) Accuracy 82.8% (95%: 81.1–84.3%) Reading 93.8% · grammar 78.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.0% · answered 99.7% · error rate 15.7% Confidence: calibration gap 12.6 points (written estimate) Cost $2.048$2,048 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 44.9 s median per request, 4 in flight | 93.8%91.6%–95.4% | 82.8% | 78.1% | 2.0%answered 100% | 12.6written estimate | $2.048$2,048at list prices | 44.916 per request4 in flight |
| Kimi K3.1 Pro no reasoning · chat‡ Open · 2.8T MoE · 104B activeother setting of Kimi K3.1 ProLatency 5.75 sCost $0.853$853 per 1,000per 1M Kimi K3.1 Pro no reasoning · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish reading 92.5% (95%: 90.3–94.3%) Accuracy 72.7% (95%: 70.8–74.5%) Reading 92.5% · grammar 64.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.9% · error rate 28.3% Confidence: calibration gap 20.1 points (written estimate) Cost $0.853$853 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.75 s median per request, 4 in flight | 92.5%90.3%–94.3% | 72.7% | 64.3% | 0.5%answered 100% | 20.1written estimate | $0.853$853at list prices | 5.7516 per request4 in flight | |
| 14 | DeepSeek-V4.1-Flash reasoning on · chat · first release‡ Open · 552B MoE · 16B activeLatency 365.1 sCost $1.299$1,299 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish reading 93.4% (95%: 91.3–95.1%) Accuracy 92.0% (95%: 90.8–93.1%) Reading 93.4% · grammar 91.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.0 points (answer letter probability) Cost $1.299$1,299 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 365.1 s median per request, 8 in flight With its dated checkpoint deepseek-v4.1-flash-1017 under its other settings | 93.4%91.3%–95.1% | 92.0% | 91.4% | Not run | 8.0answer letter probability | $1.299$1,299at a host's list price for the same weights | 365.116 per request8 in flight |
| DeepSeek-V4.1-Flash reasoning on · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 486.9 sCost $1.278$1,278 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish reading 94.1% (95%: 92.0–95.6%) Accuracy 91.8% (95%: 90.5–92.8%) Reading 94.1% · grammar 90.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.9% · error rate 7.0% Confidence: calibration gap 6.9 points (answer letter probability) Cost $1.278$1,278 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 486.9 s median per request, 8 in flight | 94.1%92.0%–95.6% | 91.8% | 90.8% | 1.3%answered 100% | 6.9answer letter probability | $1.278$1,278at the maker's list price | 486.916 per request8 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · first release‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 4.36 sCost $0.098$98.46 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish reading 87.8% (95%: 85.1–90.1%) Accuracy 65.9% (95%: 63.9–67.9%) Reading 87.8% · grammar 56.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.098$98.46 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 4.36 s median per request, 1 in flight | 87.8%85.1%–90.1% | 65.9% | 56.6% | Not run | 13.5answer letter probability | $0.098$98.46at a host's list price for the same weights | 4.3616 per request1 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 5.08 sCost $0.098$98.47 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish reading 88.3% (95%: 85.6–90.5%) Accuracy 66.0% (95%: 64.0–68.0%) Reading 88.3% · grammar 56.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 33.2% Confidence: calibration gap 16.4 points (answer letter probability) Cost $0.098$98.47 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 5.08 s median per request, 1 in flight | 88.3%85.6%–90.5% | 66.0% | 56.5% | 0.0%answered 100% | 16.4answer letter probability | $0.098$98.47at the maker's list price | 5.0816 per request1 in flight | |
| 15 | Karpo Decision 31B‡ Open · 31BLatency 0.159 sCost $0.105$105 per 1,000per 1M Karpo Decision 31B · MachinePulse · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 93.4% (95%: 91.3–95.1%) Accuracy 77.8% (95%: 76.0–79.5%) Reading 93.4% · grammar 71.1% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 23.0% Confidence: calibration gap 2.3 points (option score) Cost $0.105$105 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.159 s median per request, 1 in flight | 93.4%91.3%–95.1% | 77.8% | 71.1% | 0.6%answered 100% | 2.3option score | $0.105$105from GPU time | 0.1591 per request1 in flight |
| 16 | KnowLine-31B-Gen2‡ Open · 31BLatency 0.168 sCost $0.120$120 per 1,000per 1M KnowLine-31B-Gen2 · PelaAI · decision model Base model: Gemma 4 31B Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 93.4% (95%: 91.3–95.1%) Accuracy 81.2% (95%: 79.5–82.8%) Reading 93.4% · grammar 76.0% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 18.6% Kept at 99% confidence with no observed mistakes: 3.7% Confidence: calibration gap 5.6 points (option score) Cost $0.120$120 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.168 s median per request, 1 in flight | 93.4%91.3%–95.1% | 81.2% | 76.0% | 0.0%answered 100% | 5.6option score | $0.120$120from GPU time | 0.1681 per request1 in flight |
| 17 | Copernicus-31B‡ Open · 31BLatency 0.184 sCost $0.134$134 per 1,000per 1M Copernicus-31B · Poindexter Labs · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 93.1% (95%: 90.9–94.8%) Accuracy 81.8% (95%: 80.1–83.4%) Reading 93.1% · grammar 77.0% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.2% · answered 99.8% · error rate 18.3% Kept at 99% confidence with no observed mistakes: 4.9% Confidence: calibration gap 9.1 points (option score) Cost $0.134$134 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.184 s median per request, 1 in flight | 93.1%90.9%–94.8% | 81.8% | 77.0% | 1.2%answered 100% | 9.1option score | $0.134$134from GPU time | 0.1841 per request1 in flight |
| 18 | KnowLine-31B-Gen1.5‡ Open · 31BLatency 1.05 sCost No list price KnowLine-31B-Gen1.5 · PelaAI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 92.8% (95%: 90.6–94.6%) Accuracy 81.8% (95%: 80.1–83.4%) Reading 92.8% · grammar 77.1% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.2% · answered 100.0% · error rate 18.7% Kept at 99% confidence with no observed mistakes: 6.5% Confidence: calibration gap 3.4 points (option score) Cost: No list price · 1 question per request Latency: 1.05 s median per request, 1 in flight | 92.8%90.6%–94.6% | 81.8% | 77.1% | 0.2%answered 100% | 3.4option score | No list price | 1.051 per request1 in flight |
| 19 | Turion Jev 31B‡ Open · 31BLatency 0.405 sCost No list price Turion Jev 31B · Turion.AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 92.7% (95%: 90.4–94.4%) Accuracy 80.1% (95%: 78.4–81.7%) Reading 92.7% · grammar 74.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.3% · answered 99.3% · error rate 20.0% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 11.1 points (option score) Cost: No list price · 1 question per request Latency: 0.405 s median per request, 1 in flight | 92.7%90.4%–94.4% | 80.1% | 74.8% | 3.3%answered 99% | 11.1option score | No list price | 0.4051 per request1 in flight |
| 20 | REAX 1.3‡ Open · 57BLatency 0.250 sCost $0.185$185 per 1,000per 1M REAX 1.3 · REAX · decision model Two models run together, option probabilities averaged 50/50: 57.36B in all, all used on every question. Turkish reading 92.4% (95%: 90.1–94.2%) Accuracy 77.9% (95%: 76.2–79.6%) Reading 92.4% · grammar 71.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 4.1 points (option score) Cost $0.185$185 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 92.4%90.1%–94.2% | 77.9% | 71.8% | 0.4%answered 100% | 4.1option score | $0.185$185from GPU time | 0.2501 per request1 in flight |
| 21 | d1-ultra‡ Closed · EfficientLatency 0.223 sCost $0.014$14.34 per 1,000per 1M d1-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-ultra, read 2026-10-16) Turkish reading 92.1% (95%: 89.8–93.9%) Accuracy 77.1% (95%: 75.3–78.8%) Reading 92.1% · grammar 70.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 75.1% · answered 71.9% · error rate 13.0% Kept at 99% confidence with no observed mistakes: 4.1% Confidence: calibration gap 3.6 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.223 s median per request, 1 in flight | 92.1%89.8%–93.9% | 77.1% | 70.7% | 75.1%answered 72% | 3.6option score | $0.014$14.34at list prices | 0.2231 per request1 in flight |
| 22 | Gemini Decision 1 20261111‡ Closed · EfficientLatency 0.227 sCost $0.037$37.19 per 1,000per 1M Gemini Decision 1 20261111 · Google · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/google/gemini-decision-1, read 2026-11-11) Turkish reading 92.1% (95%: 89.8–93.9%) Accuracy 77.5% (95%: 75.7–79.2%) Reading 92.1% · grammar 71.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.5% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 14.5% Confidence: calibration gap 2.0 points (option score) Cost $0.037$37.19 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.227 s median per request, 1 in flight | 92.1%89.8%–93.9% | 77.5% | 71.3% | 1.9%answered 100% | 2.0option score | $0.037$37.19at list prices | 0.2271 per request1 in flight |
| 23 | Logits Sigma-1 Duo‡ Open · 57BLatency 0.250 sCost $0.181$181 per 1,000per 1M Logits Sigma-1 Duo · Logits · decision model Parameters: 57.36B dense, as the maker states it. Turkish reading 91.9% (95%: 89.6–93.8%) Accuracy 75.0% (95%: 73.2–76.8%) Reading 91.9% · grammar 67.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 24.7% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 5.7 points (option score) Cost $0.181$181 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 91.9%89.6%–93.8% | 75.0% | 67.8% | 0.4%answered 100% | 5.7option score | $0.181$181from GPU time | 0.2501 per request1 in flight |
| 24 | Sieve 31B v3‡ Open · 31BLatency 1.00 sCost No list price Sieve 31B v3 · Sthānika AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 91.8% (95%: 89.4–93.6%) Accuracy 74.2% (95%: 72.3–75.9%) Reading 91.8% · grammar 66.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 25.2% Kept at 99% confidence: 15.0%, 1 of them wrong Confidence: calibration gap 2.9 points (option score) Cost: No list price · 1 question per request Latency: 1.00 s median per request, 1 in flight | 91.8%89.4%–93.6% | 74.2% | 66.7% | 0.0%answered 100% | 2.9option score | No list price | 1.001 per request1 in flight |
| 25 | Qwen3.5-397B-A17B high · chat‡ Open · 397B MoE · 17B activeLatency 251.3 sCost $2.411$2,411 per 1,000per 1M Qwen3.5-397B-A17B high · chat · Alibaba · chat model Mixture of experts: 397B total, about 17B active per step. Size class uses the total (397B). Turkish reading 91.0% (95%: 88.6–93.0%) Accuracy 82.7% (95%: 81.1–84.2%) Reading 91.0% · grammar 79.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.5% · answered 99.7% · error rate 17.8% Confidence: calibration gap 6.2 points (written estimate) Cost $2.411$2,411 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 251.3 s median per request, 4 in flight | 91.0%88.6%–93.0% | 82.7% | 79.2% | 1.5%answered 100% | 6.2written estimate | $2.411$2,411at list prices | 251.316 per request4 in flight |
| 26 | Gemma 4 31B reasoning off · chat‡ Open · 31BLatency 3.97 sCost $0.048$48.22 per 1,000per 1M Gemma 4 31B reasoning off · chat · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 90.9% (95%: 88.4–92.8%) Accuracy 73.1% (95%: 71.2–74.9%) Reading 90.9% · grammar 65.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.1% · answered 98.9% · error rate 26.1% Confidence: calibration gap 24.2 points (answer letter probability) Cost $0.048$48.22 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 3.97 s median per request, 1 in flight | 90.9%88.4%–92.8% | 73.1% | 65.6% | 4.1%answered 99% | 24.2answer letter probability | $0.048$48.22at a host's list price for the same weights | 3.9716 per request1 in flight |
| Gemma 4 31B reasoning off · chat · 1 per request‡ Open · 31Bother setting of Gemma 4 31BLatency 0.298 sCost $0.068$68.19 per 1,000per 1M Gemma 4 31B reasoning off · chat · 1 per request · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish reading 89.0% (95%: 86.4–91.2%) Accuracy 66.8% (95%: 64.8–68.7%) Reading 89.0% · grammar 57.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.4% · error rate 33.5% Confidence: calibration gap 31.2 points (answer letter probability) Cost $0.068$68.19 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.298 s median per request, 1 in flight | 89.0%86.4%–91.2% | 66.8% | 57.3% | 1.9%answered 99% | 31.2answer letter probability | $0.068$68.19at a host's list price for the same weights | 0.2981 per request1 in flight | |
| 27 | GLiDE 27B‡ Open · 27BLatency 0.542 sCost No list price GLiDE 27B · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 90.9% (95%: 88.4–92.8%) Accuracy 71.1% (95%: 69.2–73.0%) Reading 90.9% · grammar 62.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.8% · answered 86.8% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 11.2% Confidence: calibration gap 2.3 points (option score) Cost: No list price · 1 question per request Latency: 0.542 s median per request, 1 in flight | 90.9%88.4%–92.8% | 71.1% | 62.7% | 39.8%answered 87% | 2.3option score | No list price | 0.5421 per request1 in flight |
| 28 | Mammoth-1‡ Open · 27BLatency 0.169 sCost $0.040$39.62 per 1,000per 1M Mammoth-1 · Base Labs · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish reading 90.4% (95%: 87.9–92.4%) Accuracy 67.1% (95%: 65.1–69.0%) Reading 90.4% · grammar 57.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 68.7% · answered 68.6% · error rate 20.8% Kept at 99% confidence with no observed mistakes: 13.0% Confidence: calibration gap 5.6 points (option score) Cost $0.040$39.62 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.169 s median per request, 1 in flight | 90.4%87.9%–92.4% | 67.1% | 57.2% | 68.7%answered 69% | 5.6option score | $0.040$39.62at the maker's list price | 0.1691 per request1 in flight |
| 29 | Muse Ember max · chat‡ Closed · EfficientLatency 30.9 sCost $0.650$650 per 1,000per 1M Muse Ember max · chat · Meta · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/meta/muse-ember, read 2026-11-10) Turkish reading 90.2% (95%: 87.7–92.3%) Accuracy 85.6% (95%: 84.1–87.0%) Reading 90.2% · grammar 83.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.1% · answered 97.5% · error rate 13.1% Confidence: calibration gap 6.2 points (written estimate) Cost $0.650$650 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 30.9 s median per request, 4 in flight | 90.2%87.7%–92.3% | 85.6% | 83.7% | 2.1%answered 98% | 6.2written estimate | $0.650$650at list prices | 30.916 per request4 in flight |
| 30 | d3-ultra‡ Open · 26BLatency 0.333 sCost $0.216$216 per 1,000per 1M d3-ultra · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.8% (95%: 87.2–91.9%) Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.8% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.5% · error rate 32.2% Kept at 99% confidence: 17.5%, 1 of them wrong Confidence: calibration gap 12.5 points (option score) Cost $0.216$216 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.333 s median per request, 1 in flight | 89.8%87.2%–91.9% | 67.8% | 58.4% | 1.4%answered 100% | 12.5option score | $0.216$216from GPU time | 0.3331 per request1 in flight |
| 31 | d1.2-ultra‡ Closed · EfficientLatency 0.220 sCost $0.014$14.34 per 1,000per 1M d1.2-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish reading 89.8% (95%: 87.2–91.9%) Accuracy 68.1% (95%: 66.1–70.0%) Reading 89.8% · grammar 58.9% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 87.3% · answered 54.7% · error rate 12.1% Kept at 99% confidence with no observed mistakes: 0.5% Confidence: calibration gap 3.4 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.220 s median per request, 1 in flight | 89.8%87.2%–91.9% | 68.1% | 58.9% | 87.3%answered 55% | 3.4option score | $0.014$14.34at list prices | 0.2201 per request1 in flight |
| 32 | GLiDE v1.2‡ Open · 27BLatency 0.413 sCost $0.014$14.37 per 1,000per 1M GLiDE v1.2 · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.6% (95%: 87.1–91.7%) Accuracy 68.5% (95%: 66.5–70.4%) Reading 89.6% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.5% · answered 85.4% · error rate 26.2% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 2.9 points (option score) Cost $0.014$14.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.413 s median per request, 1 in flight | 89.6%87.1%–91.7% | 68.5% | 59.5% | 39.5%answered 85% | 2.9option score | $0.014$14.37at the maker's list price | 0.4131 per request1 in flight |
| 33 | Drex Max‡ Open · 27BLatency 0.082 sCost $0.023$23.29 per 1,000per 1M Drex Max · Nace.AI · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.6% (95%: 87.1–91.7%) Accuracy 67.7% (95%: 65.8–69.7%) Reading 89.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.4% Kept at 99% confidence: 20.2%, 1 of them wrong Confidence: calibration gap 6.3 points (option score) Cost $0.023$23.29 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.082 s median per request, 1 in flight | 89.6%87.1%–91.7% | 67.7% | 58.4% | 0.3%answered 100% | 6.3option score | $0.023$23.29at the maker's list price | 0.0821 per request1 in flight |
| 34 | GLM-5.3-Flash reasoning on · chat‡ Open · 320B MoE · 18B activeLatency 102.6 sCost $0.543$543 per 1,000per 1M GLM-5.3-Flash reasoning on · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish reading 89.3% (95%: 86.7–91.5%) Accuracy 78.8% (95%: 77.0–80.5%) Reading 89.3% · grammar 74.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 91.9% · error rate 16.6% Confidence: calibration gap 4.5 points (written estimate) Cost $0.543$543 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 102.6 s median per request, 8 in flight | 89.3%86.7%–91.5% | 78.8% | 74.3% | 0.6%answered 92% | 4.5written estimate | $0.543$543at the maker's list price | 102.616 per request8 in flight |
| GLM-5.3-Flash reasoning off · chat‡ Open · 320B MoE · 18B activeother setting of GLM-5.3-FlashLatency 1.07 sCost $0.047$47.28 per 1,000per 1M GLM-5.3-Flash reasoning off · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish reading 82.2% (95%: 79.1–84.9%) Accuracy 57.8% (95%: 55.7–59.8%) Reading 82.2% · grammar 47.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 41.0% Confidence: calibration gap 21.0 points (written estimate) Cost $0.047$47.28 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 1.07 s median per request, 1 in flight | 82.2%79.1%–84.9% | 57.8% | 47.4% | 0.3%answered 100% | 21.0written estimate | $0.047$47.28at the maker's list price | 1.0716 per request1 in flight | |
| 35 | Qwen3.8-27B low · chat‡ Open · 27BLatency 648.4 sCost $3.095$3,095 per 1,000per 1M Qwen3.8-27B low · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish reading 89.3% (95%: 86.7–91.5%) Accuracy 74.7% (95%: 72.8–76.4%) Reading 89.3% · grammar 68.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.9% · answered 97.9% · error rate 22.9% Confidence: calibration gap 23.7 points (answer letter probability) Cost $3.095$3,095 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 648.4 s median per request, 4 in flight | 89.3%86.7%–91.5% | 74.7% | 68.4% | 5.9%answered 98% | 23.7answer letter probability | $3.095$3,095at the maker's list price | 648.416 per request4 in flight |
| Qwen3.8-27B yes/no scoring‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.280 sCost $0.704$704 per 1,000per 1M Qwen3.8-27B yes/no scoring · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish reading 80.5% (95%: 77.3–83.3%) Accuracy 51.9% (95%: 49.8–54.0%) Reading 80.5% · grammar 39.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.9 points (option score) Cost $0.704$704 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.280 s median per request, 1 in flight | 80.5%77.3%–83.3% | 51.9% | 39.8% | Not run | 8.9option score | $0.704$704at the maker's list price | 0.2801 per request1 in flighton its repeats | |
| Qwen3.8-27B stated confidence · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.433 sCost $0.234$234 per 1,000per 1M Qwen3.8-27B stated confidence · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish reading 83.4% (95%: 80.3–86.0%) Accuracy 54.2% (95%: 52.1–56.3%) Reading 83.4% · grammar 41.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.7% · error rate 45.6% Confidence: calibration gap 40.8 points (written estimate) Cost $0.234$234 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.433 s median per request, 1 in flight | 83.4%80.3%–86.0% | 54.2% | 41.8% | 0.6%answered 100% | 40.8written estimate | $0.234$234at the maker's list price | 0.4331 per request1 in flighton its repeats | |
| Qwen3.8-27B vote share · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 1.62 sCost $0.825$825 per 1,000per 1M Qwen3.8-27B vote share · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish reading 84.0% (95%: 81.0–86.6%) Accuracy 54.3% (95%: 52.2–56.4%) Reading 84.0% · grammar 41.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.4% · error rate 45.6% Confidence: calibration gap 22.5 points (agreement across ten answers) Cost $0.825$825 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 1.62 s median per request, 1 in flight | 84.0%81.0%–86.6% | 54.3% | 41.7% | 1.3%answered 99% | 22.5agreement across ten answers | $0.825$825at the maker's list price | 1.621 per request1 in flight | |
| 36 | Perplexity Decider 27B v1.2 DGX Spark‡ Open · 27BLatency 0.580 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.2 DGX Spark · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.2% (95%: 86.6–91.3%) Accuracy 67.1% (95%: 65.1–69.0%) Reading 89.2% · grammar 57.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.7% Kept at 99% confidence with no observed mistakes: 16.4% Confidence: calibration gap 4.4 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.580 s median per request, 1 in flight | 89.2%86.6%–91.3% | 67.1% | 57.7% | 0.3%answered 100% | 4.4option score | $0.008$7.76at the maker's list price | 0.5801 per request1 in flight |
| Perplexity Decider 27B v1.2 API‡ Open · 27Bother setting of Perplexity Decider 27B v1.2Latency 0.254 sCost $0.008$7.70 per 1,000per 1M Perplexity Decider 27B v1.2 API · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.0% (95%: 86.4–91.2%) Accuracy 66.3% (95%: 64.3–68.3%) Reading 89.0% · grammar 56.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 33.1% Kept at 99% confidence with no observed mistakes: 14.9% Confidence: calibration gap 4.2 points (option score) Cost $0.008$7.70 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.254 s median per request, 1 in flight | 89.0%86.4%–91.2% | 66.3% | 56.7% | 0.4%answered 100% | 4.2option score | $0.008$7.70at list prices | 0.2541 per request1 in flight | |
| 37 | d3‡ Open · 26BLatency 0.092 sCost $0.070$70.21 per 1,000per 1M d3 · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.2% (95%: 86.6–91.3%) Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.2% · grammar 58.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 32.5% Kept at 99% confidence with no observed mistakes: 11.4% Confidence: calibration gap 2.6 points (option score) Cost $0.070$70.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.092 s median per request, 1 in flight | 89.2%86.6%–91.3% | 67.8% | 58.7% | 0.0%answered 100% | 2.6option score | $0.070$70.21from GPU time | 0.0921 per request1 in flight |
| 38 | Perplexity Decider 27B v1.1‡ Open · 27BLatency 0.233 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.0% (95%: 86.4–91.2%) Accuracy 66.9% (95%: 64.9–68.8%) Reading 89.0% · grammar 57.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 15.9% Confidence: calibration gap 3.8 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.233 s median per request, 1 in flight | 89.0%86.4%–91.2% | 66.9% | 57.5% | Not run | 3.8option score | $0.008$7.76at list prices | 0.2331 per request1 in flight |
| 39 | Perplexity Decider 27B v1.3‡ Open · 27BLatency 0.110 sCost $0.081$81.18 per 1,000per 1M Perplexity Decider 27B v1.3 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 89.0% (95%: 86.4–91.2%) Accuracy 68.4% (95%: 66.5–70.3%) Reading 89.0% · grammar 59.7% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.8% · error rate 32.4% Kept at 99% confidence with no observed mistakes: 12.6% Confidence: calibration gap 5.0 points (option score) Cost $0.081$81.18 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.110 s median per request, 1 in flight | 89.0%86.4%–91.2% | 68.4% | 59.7% | 0.7%answered 100% | 5.0option score | $0.081$81.18from GPU time | 0.1101 per request1 in flight |
| – | Perplexity Decider 27B v1.2 API · with web‡ Open · 27BLatency 0.482 sCost ≤ $1.034≤ $1,034 per 1,000per 1M Perplexity Decider 27B v1.2 API · with web · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 88.9% (95%: 86.2–91.1%) Accuracy 68.7% (95%: 66.8–70.6%) Reading 88.9% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 99.0% · error rate 32.3% Kept at 99% confidence: 16.0%, 1 of them wrong Confidence: calibration gap 3.6 points (option score) Cost ≤ $1.034≤ $1,034 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.482 s median per request, 1 in flight | 88.9%86.2%–91.1% | 68.7% | 60.2% | 3.2%answered 99% | 3.6option score | ≤ $1.034≤ $1,034at list prices | 0.4821 per request1 in flight |
| 40 | Drex Max v1.1‡ Closed · Efficient (provisional)Latency 0.313 sCost $0.023$23.11 per 1,000per 1M Drex Max v1.1 · Nace.AI · decision model Parameters: not disclosed by maker Class: Efficient (provisional), by its introductory price, the only one published: $0.03 per 1M input tokens (https://nace.ai/developers, read 2026-11-10) Turkish reading 88.6% (95%: 85.9–90.8%) Accuracy 67.4% (95%: 65.4–69.3%) Reading 88.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.2% · answered 85.4% · error rate 27.5% Kept at 99% confidence with no observed mistakes: 10.0% Confidence: calibration gap 2.9 points (option score) Cost $0.023$23.11 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 88.6%85.9%–90.8% | 67.4% | 58.4% | 38.2%answered 85% | 2.9option score | $0.023$23.11at list prices | 0.3131 per request1 in flight |
| 41 | GLiDE API‡ Closed · EfficientLatency 2.09 sCost $0.160$160 per 1,000per 1M GLiDE API · Fastino · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.fastino.ai/pricing, read 2026-10-22) Turkish reading 87.8% (95%: 85.1–90.1%) Accuracy 67.9% (95%: 65.9–69.8%) Reading 87.8% · grammar 59.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 59.4% · answered 76.3% · error rate 21.2% Kept at 99% confidence with no observed mistakes: 5.5% Confidence: calibration gap 4.5 points (option score) Cost $0.160$160 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.09 s median per request, 4 in flight | 87.8%85.1%–90.1% | 67.9% | 59.4% | 59.4%answered 76% | 4.5option score | $0.160$160at list prices | 2.091 per request4 in flight |
| 42 | SubQ Snap Preview‡ Closed · EfficientLatency 0.389 sCost $0.019$19.41 per 1,000per 1M SubQ Snap Preview · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.05 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-10-13) Turkish reading 87.7% (95%: 84.9–90.0%) Accuracy 65.2% (95%: 63.2–67.2%) Reading 87.7% · grammar 55.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.9 points (option score) Cost $0.019$19.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.389 s median per request, 1 in flight | 87.7%84.9%–90.0% | 65.2% | 55.7% | Not run | 4.9option score | $0.019$19.41at list prices | 0.3891 per request1 in flight |
| 43 | GLM-5.3 low · chat‡ Open · WorkhorseLatency 17.1 sCost $0.814$814 per 1,000per 1M GLM-5.3 low · chat · Z.ai · chat model Parameters: not stated by the maker (a mixture of experts); counted from the released files: 753.33B. Class: Workhorse, by its standard list price: $1.40 per 1M input tokens (https://openrouter.ai/z-ai/glm-5.3, read 2026-10-28) Turkish reading 87.7% (95%: 84.9–90.0%) Accuracy 74.5% (95%: 72.7–76.3%) Reading 87.7% · grammar 68.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.1% · answered 99.0% · error rate 23.9% Confidence: calibration gap 2.6 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 17.1 s median per request, 4 in flight | 87.7%84.9%–90.0% | 74.5% | 68.9% | 1.1%answered 99% | 2.6written estimate | $0.814$814at list prices | 17.116 per request4 in flight |
| 44 | GPT-6 Luna low · chat‡ Closed · EfficientLatency 7.71 sCost $0.048$48.01 per 1,000per 1M GPT-6 Luna low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish reading 87.3% (95%: 84.6–89.7%) Accuracy 68.3% (95%: 66.3–70.2%) Reading 87.3% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.0% · answered 99.7% · error rate 33.2% Confidence: calibration gap 25.7 points (written estimate) Cost $0.048$48.01 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.71 s median per request, 1 in flight | 87.3%84.6%–89.7% | 68.3% | 60.2% | 1.0%answered 100% | 25.7written estimate | $0.048$48.01subscription, priced as the API | 7.7116 per request1 in flight |
| 45 | Inkling-Small high · chat · Jul 2026‡ Open · 276B MoE · 12B activeLatency 97.7 sCost $1.137$1,137 per 1,000per 1M Inkling-Small high · chat · Jul 2026 · Thinking Machines · chat model Mixture of experts: 276B total (counted from the released files: 265.96B), about 12B active per step. Size class uses the total (276B). Turkish reading 87.2% (95%: 84.4–89.5%) Accuracy 80.0% (95%: 78.3–81.6%) Reading 87.2% · grammar 76.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.1% · answered 99.3% · error rate 21.4% Confidence: calibration gap 7.9 points (written estimate) Cost $1.137$1,137 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 97.7 s median per request, 4 in flight | 87.2%84.4%–89.5% | 80.0% | 76.9% | 3.1%answered 99% | 7.9written estimate | $1.137$1,137at list prices | 97.716 per request4 in flight |
| 46 | Qwen3.8 Flash no reasoning · chat‡ Closed · EfficientLatency 10.0 sCost $0.047$47.13 per 1,000per 1M Qwen3.8 Flash no reasoning · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.15 per 1M input tokens (https://openrouter.ai/qwen/qwen3.8-flash, read 2026-11-11) Turkish reading 87.2% (95%: 84.4–89.5%) Accuracy 63.1% (95%: 61.0–65.1%) Reading 87.2% · grammar 52.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 98.3% · error rate 35.4% Confidence: calibration gap 28.7 points (written estimate) Cost $0.047$47.13 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 10.0 s median per request, 4 in flight | 87.2%84.4%–89.5% | 63.1% | 52.8% | 4.7%answered 98% | 28.7written estimate | $0.047$47.13at list prices | 10.016 per request4 in flight |
| 47 | SubQ Snap 1.1‡ Closed · EfficientLatency 0.378 sCost $0.015$15.41 per 1,000per 1M SubQ Snap 1.1 · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-11-01) Turkish reading 87.0% (95%: 84.3–89.4%) Accuracy 63.9% (95%: 61.8–65.9%) Reading 87.0% · grammar 54.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 47.1% · answered 80.1% · error rate 27.8% Kept at 99% confidence: 13.6%, 1 of them wrong Confidence: calibration gap 11.2 points (option score) Cost $0.015$15.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.378 s median per request, 1 in flight | 87.0%84.3%–89.4% | 63.9% | 54.0% | 47.1%answered 80% | 11.2option score | $0.015$15.41at list prices | 0.3781 per request1 in flight |
| 48 | Claude Haiku 5.5 max · chat‡ Closed · EfficientLatency 417.0 sCost $2.965$2,965 per 1,000per 1M Claude Haiku 5.5 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/anthropic/claude-haiku-5.5, read 2026-11-10) Turkish reading 86.9% (95%: 84.1–89.3%) Accuracy 57.1% (95%: 55.0–59.1%) Reading 86.9% · grammar 44.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.0% · answered 67.7% · error rate 9.3% Confidence: calibration gap 17.5 points (written estimate) Cost $2.965$2,965 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 417.0 s median per request, 4 in flight | 86.9%84.1%–89.3% | 57.1% | 44.4% | 22.0%answered 68% | 17.5written estimate | $2.965$2,965at list prices | 417.016 per request4 in flight |
| 49 | MiMo-V2.6-Flash high · chat‡ Open · 309B MoE · 15B activeLatency 21.4 sCost $2.219$2,219 per 1,000per 1M MiMo-V2.6-Flash high · chat · Xiaomi · chat model Mixture of experts: 309B total, about 15B active per step. Size class uses the total (309B). Turkish reading 86.6% (95%: 83.8–89.0%) Accuracy 72.1% (95%: 70.2–73.9%) Reading 86.6% · grammar 66.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.5% · answered 96.7% · error rate 26.2% Confidence: calibration gap 7.1 points (written estimate) Cost $2.219$2,219 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 21.4 s median per request, 32 in flight | 86.6%83.8%–89.0% | 72.1% | 66.0% | 3.5%answered 97% | 7.1written estimate | $2.219$2,219at list prices | 21.41 per request32 in flight |
| 50 | Perplexity Decider 27B v1‡ Open · 27BLatency 0.216 sCost $0.016$15.54 per 1,000per 1M Perplexity Decider 27B v1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish reading 86.0% (95%: 83.1–88.4%) Accuracy 60.4% (95%: 58.3–62.4%) Reading 86.0% · grammar 49.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.0% Confidence: calibration gap 5.0 points (option score) Cost $0.016$15.54 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.216 s median per request, 1 in flight | 86.0%83.1%–88.4% | 60.4% | 49.5% | Not run | 5.0option score | $0.016$15.54at list prices | 0.2161 per request1 in flighton its repeats |
| 51 | KnowLine-12B-Gen1‡ Open · 12BLatency 0.071 sCost $0.052$52.28 per 1,000per 1M KnowLine-12B-Gen1 · PelaAI · decision model Base model: Gemma 4 12B Unified Parameters: 12B dense, as the maker states it (counted from the released files: 11.96B). Turkish reading 86.0% (95%: 83.1–88.4%) Accuracy 62.5% (95%: 60.4–64.5%) Reading 86.0% · grammar 52.5% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 36.6% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.2 points (option score) Cost $0.052$52.28 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.071 s median per request, 1 in flight | 86.0%83.1%–88.4% | 62.5% | 52.5% | 0.1%answered 100% | 6.2option score | $0.052$52.28from GPU time | 0.0711 per request1 in flight |
| 52 | d1.2‡ Closed · EfficientLatency 0.213 sCost $0.007$6.83 per 1,000per 1M d1.2 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish reading 86.0% (95%: 83.1–88.4%) Accuracy 61.7% (95%: 59.7–63.7%) Reading 86.0% · grammar 51.4% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 79.4% · answered 55.7% · error rate 20.7% Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 3.3 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.213 s median per request, 1 in flight | 86.0%83.1%–88.4% | 61.7% | 51.4% | 79.4%answered 56% | 3.3option score | $0.007$6.83at list prices | 0.2131 per request1 in flight |
| 53 | GPT-6 Luna Decisions API‡ Closed · EfficientLatency 0.191 sCost $0.040$40.44 per 1,000per 1M GPT-6 Luna Decisions API · OpenAI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish reading 85.5% (95%: 82.6–88.0%) Accuracy 63.6% (95%: 61.6–65.6%) Reading 85.5% · grammar 54.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 16.7%, 3 of them wrong Confidence: calibration gap 2.4 points (option score) Cost $0.040$40.44 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.191 s median per request, 4 in flight | 85.5%82.6%–88.0% | 63.6% | 54.3% | Not run | 2.4option score | $0.040$40.44at list prices | 0.1911 per request4 in flight |
| 54 | Claude Sonnet 5.5 high · chat‡ Closed · WorkhorseLatency 20.5 sCost $3.654$3,654 per 1,000per 1M Claude Sonnet 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 85.2% (95%: 82.3–87.7%) Accuracy 85.0% (95%: 83.5–86.5%) Reading 85.2% · grammar 85.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.9% · error rate 12.8% Confidence: calibration gap 16.0 points (written estimate) Cost $3.654$3,654 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 20.5 s median per request, 1 in flight | 85.2%82.3%–87.7% | 85.0% | 85.0% | 0.7%answered 100% | 16.0written estimate | $3.654$3,654subscription, priced as the API | 20.516 per request1 in flighton a sample |
| Claude Sonnet 5.5 high · chat · 1 per request‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 4.64 sCost $7.748$7,748 per 1,000per 1M Claude Sonnet 5.5 high · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 94.2% (95%: 92.1–95.8%) Accuracy 89.2% (95%: 87.8–90.4%) Reading 94.2% · grammar 87.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.9% · error rate 11.1% Confidence: calibration gap 9.8 points (written estimate) Cost $7.748$7,748 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 4.64 s median per request, 1 in flight | 94.2%92.1%–95.8% | 89.2% | 87.0% | 0.8%answered 100% | 9.8written estimate | $7.748$7,748subscription, priced as the API | 4.641 per request1 in flight | |
| Claude Sonnet 5.5 low · chat‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 3.87 sCost $1.896$1,896 per 1,000per 1M Claude Sonnet 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish reading 82.9% (95%: 79.9–85.6%) Accuracy 61.4% (95%: 59.3–63.4%) Reading 82.9% · grammar 52.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.8% · error rate 35.1% Confidence: calibration gap 7.9 points (written estimate) Cost $1.896$1,896 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 3.87 s median per request, 1 in flight | 82.9%79.9%–85.6% | 61.4% | 52.2% | 0.6%answered 100% | 7.9written estimate | $1.896$1,896subscription, priced as the API | 3.8716 per request1 in flighton a sample | |
| 55 | celeris-1-decision‡ Closed · EfficientLatency 0.401 sCost $0.015$15.28 per 1,000per 1M celeris-1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-10-10) Turkish reading 85.2% (95%: 82.3–87.7%) Accuracy 60.5% (95%: 58.4–62.5%) Reading 85.2% · grammar 49.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.3 points (option score) Cost $0.015$15.28 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.401 s median per request, 1 in flight | 85.2%82.3%–87.7% | 60.5% | 49.9% | Not run | 4.3option score | $0.015$15.28at list prices | 0.4011 per request1 in flight |
| 56 | OneJev-27B‡ Open · 27BLatency 0.125 sCost $0.099$99.09 per 1,000per 1M OneJev-27B · OmniJev · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish reading 85.2% (95%: 82.3–87.7%) Accuracy 59.6% (95%: 57.5–61.6%) Reading 85.2% · grammar 48.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 40.6% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 5.5 points (option score) Cost $0.099$99.09 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.125 s median per request, 1 in flight | 85.2%82.3%–87.7% | 59.6% | 48.7% | 0.1%answered 100% | 5.5option score | $0.099$99.09from GPU time | 0.1251 per request1 in flight |
| 57 | Clef‡ Open · 27BLatency 0.205 sCost $0.110$110 per 1,000per 1M Clef · Cloudflare · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.48B). Turkish reading 85.1% (95%: 82.1–87.6%) Accuracy 57.3% (95%: 55.2–59.3%) Reading 85.1% · grammar 45.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.4% · answered 96.8% · error rate 42.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.7 points (option score) Cost $0.110$110 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.205 s median per request, 1 in flight | 85.1%82.1%–87.6% | 57.3% | 45.5% | 7.4%answered 97% | 10.7option score | $0.110$110at the maker's list price | 0.2051 per request1 in flight |
| 58 | Jev 1.14‡ Closed · EfficientLatency 0.211 sCost $0.029$28.84 per 1,000per 1M Jev 1.14 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://docs.typesafe.ai/models, read 2026-10-16) Turkish reading 85.1% (95%: 82.1–87.6%) Accuracy 59.2% (95%: 57.1–61.2%) Reading 85.1% · grammar 48.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 11.8%, 1 of them wrong Confidence: calibration gap 2.5 points (option score) Cost $0.029$28.84 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.211 s median per request, 1 in flight | 85.1%82.1%–87.6% | 59.2% | 48.2% | Not run | 2.5option score | $0.029$28.84at list prices | 0.2111 per request1 in flight |
| 59 | d1.1‡ Closed · EfficientLatency 0.195 sCost $0.007$6.83 per 1,000per 1M d1.1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-24) Turkish reading 84.6% (95%: 81.6–87.2%) Accuracy 61.5% (95%: 59.5–63.5%) Reading 84.6% · grammar 51.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 4.0% Confidence: calibration gap 2.0 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.195 s median per request, 1 in flight | 84.6%81.6%–87.2% | 61.5% | 51.7% | Not run | 2.0option score | $0.007$6.83at list prices | 0.1951 per request1 in flight |
| 60 | Mercury Decide‡ Closed · EfficientLatency 0.313 sCost $0.008$7.71 per 1,000per 1M Mercury Decide · Inception · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://www.inceptionlabs.ai/, read 2026-11-10) Turkish reading 84.5% (95%: 81.5–87.0%) Accuracy 58.3% (95%: 56.2–60.3%) Reading 84.5% · grammar 47.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 20.5%, 13 of them wrong Confidence: calibration gap 15.5 points (option score) Cost $0.008$7.71 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 84.5%81.5%–87.0% | 58.3% | 47.1% | Not run | 15.5option score | $0.008$7.71at list prices | 0.3131 per request1 in flight |
| 61 | Jev 1.13‡ Closed · EfficientLatency 0.243 sCost $0.029$28.85 per 1,000per 1M Jev 1.13 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://typesafe.ai/blog/introducing-system-one-models-and-jev, read 2026-10-08) Turkish reading 83.7% (95%: 80.7–86.3%) Accuracy 57.5% (95%: 55.4–59.6%) Reading 83.7% · grammar 46.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 14.2%, 1 of them wrong Confidence: calibration gap 5.1 points (option score) Cost $0.029$28.85 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.243 s median per request, 1 in flight | 83.7%80.7%–86.3% | 57.5% | 46.4% | Not run | 5.1option score | $0.029$28.85at list prices | 0.2431 per request1 in flighton its repeats |
| 62 | Open-Jev-27B-v1.1‡ Open · 27BLatency 0.386 sCost $0.271$271 per 1,000per 1M Open-Jev-27B-v1.1 · Zefan Cai · decision model Base model: Qwen3.8-27B + adapter Parameters: 27B dense, as the maker states it (counted from the released files: 27.80B). Turkish reading 82.9% (95%: 79.9–85.6%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.9% · grammar 44.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.4% · error rate 44.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 24.5 points (option score) Cost $0.271$271 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.386 s median per request, 1 in flight | 82.9%79.9%–85.6% | 55.7% | 44.1% | 1.4%answered 99% | 24.5option score | $0.271$271from GPU time | 0.3861 per request1 in flighton its repeats |
| 63 | MiniMax-M3.1-Flash high · chat‡ Open · 428B MoE · 23B activeLatency 112.6 sCost $0.814$814 per 1,000per 1M MiniMax-M3.1-Flash high · chat · MiniMax · chat model Mixture of experts: 428B total, about 23B active per step. Size class uses the total (428B). Turkish reading 82.9% (95%: 79.9–85.6%) Accuracy 66.5% (95%: 64.5–68.4%) Reading 82.9% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.5% · answered 97.6% · error rate 32.8% Confidence: calibration gap 21.1 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 112.6 s median per request, 4 in flight | 82.9%79.9%–85.6% | 66.5% | 59.5% | 5.5%answered 98% | 21.1written estimate | $0.814$814at list prices | 112.616 per request4 in flight |
| 64 | Jeffan-9B-1211‡ Open · 9.0BLatency 0.064 sCost $0.044$44.29 per 1,000per 1M Jeffan-9B-1211 · tattrongvu · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish reading 82.9% (95%: 79.9–85.6%) Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.9% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.1% · error rate 42.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.6 points (option score) Cost $0.044$44.29 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.064 s median per request, 1 in flight | 82.9%79.9%–85.6% | 57.0% | 46.0% | 4.3%answered 98% | 4.6option score | $0.044$44.29from GPU time | 0.0641 per request1 in flight |
| 65 | d1‡ Closed · EfficientLatency 0.305 sCost $0.007$7.43 per 1,000per 1M d1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-18) Turkish reading 82.8% (95%: 79.7–85.5%) Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.8% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 12.5%, 1 of them wrong Confidence: calibration gap 10.2 points (option score) Cost $0.007$7.43 per 1,000 questionsper 1M questions (free tier, at the list price) · 1 question per request Latency: 0.305 s median per request, 1 in flight | 82.8%79.7%–85.5% | 57.0% | 46.0% | Not run | 10.2option score | $0.007$7.43free tier, at the list price | 0.3051 per request1 in flighton its repeats |
| 66 | Cygnet‡ Open · 12BLatency 0.036 sCost $0.039$39.39 per 1,000per 1M Cygnet · Blockbrain Labs · decision model Base model: Gemma 4 12B IT Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Turkish reading 82.8% (95%: 79.7–85.5%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.8% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 98.5% · error rate 44.2% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 13.8 points (option score) Cost $0.039$39.39 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.036 s median per request, 1 in flight | 82.8%79.7%–85.5% | 55.7% | 44.2% | 3.2%answered 99% | 13.8option score | $0.039$39.39at a host's list price for the same weights | 0.0361 per request1 in flight |
| 67 | Winnow-12B‡ Open · 12BLatency 0.059 sCost $0.041$41.40 per 1,000per 1M Winnow-12B · EldanRing · decision model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.91B). Turkish reading 82.0% (95%: 78.9–84.8%) Accuracy 55.5% (95%: 53.4–57.5%) Reading 82.0% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.0% · error rate 44.8% Kept at 99% confidence: 7.4%, 1 of them wrong Confidence: calibration gap 19.7 points (option score) Cost $0.041$41.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 82.0%78.9%–84.8% | 55.5% | 44.2% | 4.3%answered 98% | 19.7option score | $0.041$41.40from GPU time | 0.0591 per request1 in flight |
| 68 | Drex 2.0‡ Open · 35B MoE · 3.0B activeLatency 0.325 sCost $0.019$19.15 per 1,000per 1M Drex 2.0 · Nace.AI · decision model Mixture of experts: 35B total (counted from the released files: 35.11B), about 3B active per step. Size class uses the total (35B). Turkish reading 81.4% (95%: 78.2–84.2%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 81.4% · grammar 44.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 4.2 points (option score) Cost $0.019$19.15 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.325 s median per request, 1 in flight | 81.4%78.2%–84.2% | 55.7% | 44.8% | Not run | 4.2option score | $0.019$19.15at list prices | 0.3251 per request1 in flight |
| 69 | Adjudex 9B v5‡ Open · 9.0BLatency 0.038 sCost $0.026$26.30 per 1,000per 1M Adjudex 9B v5 · newsbang · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish reading 80.8% (95%: 77.6–83.6%) Accuracy 53.1% (95%: 51.0–55.2%) Reading 80.8% · grammar 41.4% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.6% · error rate 46.6% Kept at 99% confidence: 6.0%, 2 of them wrong Confidence: calibration gap 9.2 points (option score) Cost $0.026$26.30 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.038 s median per request, 1 in flight | 80.8%77.6%–83.6% | 53.1% | 41.4% | 0.8%answered 100% | 9.2option score | $0.026$26.30from GPU time | 0.0381 per request1 in flight |
| 70 | Elephant-1‡ Open · 9.0BLatency 0.074 sCost $0.014$13.87 per 1,000per 1M Elephant-1 · Base Labs · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.41B). Turkish reading 80.5% (95%: 77.3–83.3%) Accuracy 52.4% (95%: 50.3–54.4%) Reading 80.5% · grammar 40.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 34.8% · answered 81.6% · error rate 42.3% Kept at 99% confidence with no observed mistakes: 7.1% Confidence: calibration gap 11.8 points (option score) Cost $0.014$13.87 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.074 s median per request, 1 in flight | 80.5%77.3%–83.3% | 52.4% | 40.4% | 34.8%answered 82% | 11.8option score | $0.014$13.87at the maker's list price | 0.0741 per request1 in flight |
| 71 | GLiDE 9B‡ Open · 7.9BLatency 0.207 sCost No list price GLiDE 9B · Fastino · decision model Parameters: 7.94B dense, as the maker states it (counted from the released files: 7.94B). Turkish reading 80.2% (95%: 77.0–83.1%) Accuracy 52.0% (95%: 50.0–54.1%) Reading 80.2% · grammar 40.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 24.1% · answered 87.4% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 8.5 points (option score) Cost: No list price · 1 question per request Latency: 0.207 s median per request, 1 in flight | 80.2%77.0%–83.1% | 52.0% | 40.1% | 24.1%answered 87% | 8.5option score | No list price | 0.2071 per request1 in flight |
| 72 | KnowLine-9B-Gen1‡ Open · 9.0BLatency 0.057 sCost $0.041$40.69 per 1,000per 1M KnowLine-9B-Gen1 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish reading 79.9% (95%: 76.6–82.8%) Accuracy 54.5% (95%: 52.5–56.6%) Reading 79.9% · grammar 43.8% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.8% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 2.8% Confidence: calibration gap 4.6 points (option score) Cost $0.041$40.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.057 s median per request, 1 in flight | 79.9%76.6%–82.8% | 54.5% | 43.8% | 0.5%answered 100% | 4.6option score | $0.041$40.69from GPU time | 0.0571 per request1 in flight |
| 73 | Cerno 9B v1.2‡ Open · 9.0BLatency 0.051 sCost $0.036$36.02 per 1,000per 1M Cerno 9B v1.2 · Apertix AI · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish reading 79.7% (95%: 76.5–82.6%) Accuracy 54.5% (95%: 52.4–56.5%) Reading 79.7% · grammar 43.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.2% · answered 99.9% · error rate 45.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.8 points (option score) Cost $0.036$36.02 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.051 s median per request, 1 in flight | 79.7%76.5%–82.6% | 54.5% | 43.7% | 0.2%answered 100% | 9.8option score | $0.036$36.02from GPU time | 0.0511 per request1 in flight |
| 74 | Dexy-Decision-9B-V2‡ Open · 9.0BLatency 0.031 sCost $0.023$23.03 per 1,000per 1M Dexy-Decision-9B-V2 · Patrick-18 · decision model Base model: vllm-sr/d3-flash Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish reading 79.6% (95%: 76.3–82.5%) Accuracy 51.6% (95%: 49.5–53.7%) Reading 79.6% · grammar 39.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.4% · answered 98.3% · error rate 48.4% Kept at 99% confidence: 6.3%, 1 of them wrong Confidence: calibration gap 9.1 points (option score) Cost $0.023$23.03 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.031 s median per request, 1 in flight | 79.6%76.3%–82.5% | 51.6% | 39.7% | 3.4%answered 98% | 9.1option score | $0.023$23.03from GPU time | 0.0311 per request1 in flight |
| 75 | KnowLine-9B-Gen2‡ Open · 9.0BLatency 0.048 sCost $0.035$35.40 per 1,000per 1M KnowLine-9B-Gen2 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish reading 79.6% (95%: 76.3–82.5%) Accuracy 53.3% (95%: 51.2–55.4%) Reading 79.6% · grammar 42.1% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 2.6% · answered 98.8% · error rate 47.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.1 points (option score) Cost $0.035$35.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.048 s median per request, 1 in flight | 79.6%76.3%–82.5% | 53.3% | 42.1% | 2.6%answered 99% | 5.1option score | $0.035$35.40from GPU time | 0.0481 per request1 in flight |
| 76 | Qwen3.5-4B reasoning on · chat‡ Open · 4.0BLatency 140.3 sCost $2.998$2,998 per 1,000per 1M Qwen3.5-4B reasoning on · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish reading 79.1% (95%: 75.8–82.1%) Accuracy 58.9% (95%: 56.8–61.0%) Reading 79.1% · grammar 50.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.3% · answered 98.4% · error rate 40.7% Confidence: not measured Cost $2.998$2,998 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 140.3 s median per request, 32 in flight | 79.1%75.8%–82.1% | 58.9% | 50.3% | 3.3%answered 98% | Not measureda plain answer line, no confidence | $2.998$2,998from GPU time | 140.31 per request32 in flight |
| 77 | StartLux-Decision-v1.1-9B-Instant‡ Open · 9.0BLatency 0.154 sCost No list price StartLux-Decision-v1.1-9B-Instant · StartLux · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish reading 78.8% (95%: 75.5–81.8%) Accuracy 51.2% (95%: 49.1–53.3%) Reading 78.8% · grammar 39.5% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 2.2% · answered 98.9% · error rate 49.2% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 7.6 points (option score) Cost: No list price · 1 question per request Latency: 0.154 s median per request, 1 in flight | 78.8%75.5%–81.8% | 51.2% | 39.5% | 2.2%answered 99% | 7.6option score | No list price | 0.1541 per request1 in flight |
| 78 | Gemma 4 E4B default reasoning · chat‡ Open · 8.0BLatency 13.4 sCost $0.307$307 per 1,000per 1M Gemma 4 E4B default reasoning · chat · Google · chat model Parameters: 8B dense · 4.5B effective, as the maker states it (counted from the released files: 8.00B). Turkish reading 77.6% (95%: 74.2–80.6%) Accuracy 57.6% (95%: 55.5–59.6%) Reading 77.6% · grammar 49.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.7% · answered 86.0% · error rate 38.4% Confidence: not measured Cost $0.307$307 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 13.4 s median per request, 32 in flight | 77.6%74.2%–80.6% | 57.6% | 49.0% | 29.7%answered 86% | Not measureda plain answer line, no confidence | $0.307$307from GPU time | 13.41 per request32 in flight |
| 79 | vera-medium-omni‡ Open · 4.2BLatency 0.029 sCost $0.021$20.72 per 1,000per 1M vera-medium-omni · scar-ai · decision model Base model: Qwen3.5-4B Parameters: 4.2B dense, as the maker states it (counted from the released files: 4.21B). Turkish reading 75.5% (95%: 72.0–78.6%) Accuracy 49.1% (95%: 47.0–51.2%) Reading 75.5% · grammar 37.9% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 8.9% · answered 95.2% · error rate 51.4% Kept at 99% confidence with no observed mistakes: 0.2% Confidence: calibration gap 3.7 points (option score) Cost $0.021$20.72 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.029 s median per request, 1 in flight | 75.5%72.0%–78.6% | 49.1% | 37.9% | 8.9%answered 95% | 3.7option score | $0.021$20.72from GPU time | 0.0291 per request1 in flight |
| 80 | Valen-Decision-1.5-9B‡ Open · 9.0BLatency 0.059 sCost $0.040$40.49 per 1,000per 1M Valen-Decision-1.5-9B · Valen-Team · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.40B). Turkish reading 75.0% (95%: 71.5–78.2%) Accuracy 47.0% (95%: 44.9–49.0%) Reading 75.0% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.3% Confidence: calibration gap 3.8 points (option score) Cost $0.040$40.49 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 75.0%71.5%–78.2% | 47.0% | 35.0% | 0.1%answered 100% | 3.8option score | $0.040$40.49from GPU time | 0.0591 per request1 in flight |
| 81 | KnowLine 4B Gen5.5‡ Open · 4.0BLatency 0.099 sCost No list price KnowLine 4B Gen5.5 · PelaAI · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish reading 74.8% (95%: 71.4–78.0%) Accuracy 48.5% (95%: 46.4–50.6%) Reading 74.8% · grammar 37.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 97.5% · error rate 52.3% Kept at 99% confidence with no observed mistakes: 1.8% Confidence: calibration gap 13.5 points (option score) Cost: No list price · 1 question per request Latency: 0.099 s median per request, 1 in flight | 74.8%71.4%–78.0% | 48.5% | 37.3% | 4.7%answered 97% | 13.5option score | No list price | 0.0991 per request1 in flight |
| 82 | Perplexity Decider Flash v1.3‡ Open · 4.0BLatency 0.033 sCost $0.025$24.69 per 1,000per 1M Perplexity Decider Flash v1.3 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish reading 74.4% (95%: 70.9–77.6%) Accuracy 49.7% (95%: 47.6–51.8%) Reading 74.4% · grammar 39.2% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 6.5% · answered 96.6% · error rate 50.4% Kept at 99% confidence with no observed mistakes: 1.3% Confidence: calibration gap 12.6 points (option score) Cost $0.025$24.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.033 s median per request, 1 in flight | 74.4%70.9%–77.6% | 49.7% | 39.2% | 6.5%answered 97% | 12.6option score | $0.025$24.69from GPU time | 0.0331 per request1 in flight |
| 83 | StartLux-Decision-v1.1-4B-Instant‡ Open · 4.0BLatency 0.074 sCost No list price StartLux-Decision-v1.1-4B-Instant · StartLux · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish reading 74.1% (95%: 70.6–77.3%) Accuracy 47.9% (95%: 45.8–50.0%) Reading 74.1% · grammar 36.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.4% Confidence: calibration gap 7.1 points (option score) Cost: No list price · 1 question per request Latency: 0.074 s median per request, 1 in flight | 74.1%70.6%–77.3% | 47.9% | 36.7% | 1.8%answered 99% | 7.1option score | No list price | 0.0741 per request1 in flight |
| 84 | Microsoft-Decision-1‡ Closed · EfficientLatency 0.286 sCost $0.014$13.82 per 1,000per 1M Microsoft-Decision-1 · Microsoft · decision model Base model: Qwen3.5-9B Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://openrouter.ai/microsoft/microsoft-decision-1, read 2026-10-17) Turkish reading 73.5% (95%: 70.0–76.7%) Accuracy 46.5% (95%: 44.4–48.5%) Reading 73.5% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 2.8%, 3 of them wrong Confidence: calibration gap 19.8 points (option score) Cost $0.014$13.82 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.286 s median per request, 1 in flight | 73.5%70.0%–76.7% | 46.5% | 35.0% | Not run | 19.8option score | $0.014$13.82at list prices | 0.2861 per request1 in flight |
| 85 | Clef-Flash‡ Open · 9.0BLatency 0.075 sCost $0.017$17.37 per 1,000per 1M Clef-Flash · Cloudflare · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.53B). Turkish reading 73.0% (95%: 69.5–76.3%) Accuracy 45.7% (95%: 43.6–47.8%) Reading 73.0% · grammar 34.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.0% · answered 95.0% · error rate 53.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 15.5 points (option score) Cost $0.017$17.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.075 s median per request, 1 in flight | 73.0%69.5%–76.3% | 45.7% | 34.1% | 9.0%answered 95% | 15.5option score | $0.017$17.37at the maker's list price | 0.0751 per request1 in flight |
| 86 | Perplexity Decider Flash v1.2‡ Open · 4.0BLatency 0.132 sCost $0.004$3.88 per 1,000per 1M Perplexity Decider Flash v1.2 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish reading 72.7% (95%: 69.2–76.0%) Accuracy 47.8% (95%: 45.7–49.9%) Reading 72.7% · grammar 37.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.7% · answered 91.8% · error rate 51.7% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost $0.004$3.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.132 s median per request, 1 in flight | 72.7%69.2%–76.0% | 47.8% | 37.2% | 14.7%answered 92% | 8.3option score | $0.004$3.88at the maker's list price | 0.1321 per request1 in flight |
| 87 | Erk-14B reasoning off · chat Open · 14BLatency 2.39 sCost $0.106$106 per 1,000per 1M Erk-14B reasoning off · chat · eCloud Technology · chat model Parameters: 14B dense, as the maker states it (counted from the released files: 14.77B). Turkish reading 72.3% (95%: 68.7–75.5%) Accuracy 45.0% (95%: 42.9–47.1%) Reading 72.3% · grammar 33.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.8% · error rate 55.3% Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.106$106 per 1,000 questionsper 1M questions (from GPU time) · 16 questions per request Latency: 2.39 s median per request, 1 in flight | 72.3%68.7%–75.5% | 45.0% | 33.4% | 0.3%answered 100% | 13.5answer letter probability | $0.106$106from GPU time | 2.3916 per request1 in flight |
| 88 | Qwen3-14B reasoning off · chat‡ Open · 15BLatency 2.85 sCost $0.032$32.31 per 1,000per 1M Qwen3-14B reasoning off · chat · Alibaba · chat model Parameters: 14.8B dense, as the maker states it (counted from the released files: 14.77B). Turkish reading 72.1% (95%: 68.6–75.4%) Accuracy 45.7% (95%: 43.6–47.8%) Reading 72.1% · grammar 34.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.0% · error rate 53.8% Confidence: calibration gap 41.0 points (answer letter probability) Cost $0.032$32.31 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 2.85 s median per request, 1 in flight | 72.1%68.6%–75.4% | 45.7% | 34.5% | 1.9%answered 99% | 41.0answer letter probability | $0.032$32.31at a host's list price for the same weights | 2.8516 per request1 in flight |
| 89 | Verdin-4B-v1.1‡ Open · 4.0BLatency 0.049 sCost $0.034$34.21 per 1,000per 1M Verdin-4B-v1.1 · victorbennett · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish reading 71.2% (95%: 67.6–74.5%) Accuracy 46.8% (95%: 44.7–48.9%) Reading 71.2% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.8% · error rate 53.3% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 9.8 points (option score) Cost $0.034$34.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.049 s median per request, 1 in flight | 71.2%67.6%–74.5% | 46.8% | 36.4% | 0.4%answered 100% | 9.8option score | $0.034$34.21from GPU time | 0.0491 per request1 in flight |
| 90 | Chinese Decision Pointer 4B‡ Open · 4.0BLatency 0.035 sCost $0.025$24.58 per 1,000per 1M Chinese Decision Pointer 4B · quincyqiang · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish reading 70.9% (95%: 67.3–74.2%) Accuracy 46.7% (95%: 44.6–48.8%) Reading 70.9% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 11.2% · answered 93.8% · error rate 52.5% Kept at 99% confidence: 2.3%, 1 of them wrong Confidence: calibration gap 11.7 points (option score) Cost $0.025$24.58 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.035 s median per request, 1 in flight | 70.9%67.3%–74.2% | 46.7% | 36.4% | 11.2%answered 94% | 11.7option score | $0.025$24.58from GPU time | 0.0351 per request1 in flight |
| 91 | Qwen3.5-4B reasoning off · chat‡ Open · 4.0BLatency 32.2 sCost $0.894$894 per 1,000per 1M Qwen3.5-4B reasoning off · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish reading 68.4% (95%: 64.8–71.9%) Accuracy 46.2% (95%: 44.1–48.3%) Reading 68.4% · grammar 36.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.9% · answered 95.4% · error rate 54.0% Confidence: not measured Cost $0.894$894 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 32.2 s median per request, 32 in flight | 68.4%64.8%–71.9% | 46.2% | 36.7% | 6.9%answered 95% | Not measureda plain answer line, no confidence | $0.894$894from GPU time | 32.21 per request32 in flight |
| 92 | Metask-Jev-4B Open · 4.5BLatency 0.090 sCost $0.063$62.82 per 1,000per 1M Metask-Jev-4B · Metask Lab · decision model Parameters: 4.54B dense, as the maker states it (counted from the released files: 4.54B). Turkish reading 62.8% (95%: 59.0–66.4%) Accuracy 40.8% (95%: 38.7–42.8%) Reading 62.8% · grammar 31.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.5% · error rate 60.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.063$62.82 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.090 s median per request, 1 in flight | 62.8%59.0%–66.4% | 40.8% | 31.4% | 0.8%answered 100% | 5.8option score | $0.063$62.82from GPU time | 0.0901 per request1 in flight |
| 93 | KnowLine-2B-Gen1‡ Open · 2.0BLatency 0.062 sCost No list price KnowLine-2B-Gen1 · PelaAI · decision model Base model: Qwen3.5-2B Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish reading 54.3% (95%: 50.4–58.0%) Accuracy 36.8% (95%: 34.8–38.8%) Reading 54.3% · grammar 29.4% ‡ Turkish not officially supported: it lists other languages (source, read 13 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.4 points (option score) Cost: No list price · 1 question per request Latency: 0.062 s median per request, 1 in flight | 54.3%50.4%–58.0% | 36.8% | 29.4% | Not run | 8.4option score | No list price | 0.0621 per request1 in flight |
| 94 | StartLux-Decision-v1.1-2B-Instant‡ Open · 2.0BLatency 0.040 sCost No list price StartLux-Decision-v1.1-2B-Instant · StartLux · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish reading 52.9% (95%: 49.1–56.7%) Accuracy 34.3% (95%: 32.3–36.3%) Reading 52.9% · grammar 26.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 21.5% · answered 84.8% · error rate 65.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.4 points (option score) Cost: No list price · 1 question per request Latency: 0.040 s median per request, 1 in flight | 52.9%49.1%–56.7% | 34.3% | 26.3% | 21.5%answered 85% | 6.4option score | No list price | 0.0401 per request1 in flight |
| 95 | GLiDE 2B‡ Open · 1.9BLatency 0.034 sCost No list price GLiDE 2B · Fastino · decision model Parameters: 1.88B dense, as the maker states it (counted from the released files: 1.88B). Turkish reading 51.2% (95%: 47.4–55.0%) Accuracy 35.3% (95%: 33.3–37.3%) Reading 51.2% · grammar 28.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.5% · answered 72.2% · error rate 61.7% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 20.7 points (option score) Cost: No list price · 1 question per request Latency: 0.034 s median per request, 1 in flight | 51.2%47.4%–55.0% | 35.3% | 28.5% | 38.5%answered 72% | 20.7option score | No list price | 0.0341 per request1 in flight |
| 96 | LFM2.5-2.6B default reasoning · chat‡ Open · 2.7BLatency 51.2 sCost $2.761$2,761 per 1,000per 1M LFM2.5-2.6B default reasoning · chat · Liquid AI · chat model Parameters: 2.69B dense, as the maker states it (counted from the released files: 2.70B). Turkish reading 45.4% (95%: 41.7–49.3%) Accuracy 33.9% (95%: 31.9–35.9%) Reading 45.4% · grammar 29.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 24.7% · answered 81.6% · error rate 63.2% Confidence: not measured Cost $2.761$2,761 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 51.2 s median per request, 32 in flight | 45.4%41.7%–49.3% | 33.9% | 29.0% | 24.7%answered 82% | Not measureda plain answer line, no confidence | $2.761$2,761at the maker's list price | 51.21 per request32 in flight |
| 97 | LFM3-1.7B reasoning on · chat‡ Open · 1.7BLatency 24.2 sCost $1.277$1,277 per 1,000per 1M LFM3-1.7B reasoning on · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish reading 44.2% (95%: 40.5–48.0%) Accuracy 32.4% (95%: 30.5–34.4%) Reading 44.2% · grammar 27.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 67.0% · answered 45.5% · error rate 58.8% Confidence: not measured Cost $1.277$1,277 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 24.2 s median per request, 32 in flight | 44.2%40.5%–48.0% | 32.4% | 27.4% | 67.0%answered 46% | Not measureda plain answer line, no confidence | $1.277$1,277at the maker's list price | 24.21 per request32 in flight |
| 98 | GLiDE 0.8B‡ Open · 0.8BLatency 0.022 sCost No list price GLiDE 0.8B · Fastino · decision model Parameters: 750M dense, as the maker states it (counted from the released files: 0.75B). Turkish reading 42.4% (95%: 38.7–46.2%) Accuracy 29.3% (95%: 27.4–31.2%) Reading 42.4% · grammar 23.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.2% · answered 78.2% · error rate 67.7% Kept at 99% confidence: 0.8%, 1 of them wrong Confidence: calibration gap 24.8 points (option score) Cost: No list price · 1 question per request Latency: 0.022 s median per request, 1 in flight | 42.4%38.7%–46.2% | 29.3% | 23.7% | 29.2%answered 78% | 24.8option score | No list price | 0.0221 per request1 in flight |
| 99 | d1-2B‡ Open · 1.9BLatency 0.019 sCost $0.013$13.17 per 1,000per 1M d1-2B · Liquid AI · decision model Base model: LFM3-VL-2B Parameters: 1.92B dense, as the maker states it (counted from the released files: 1.92B). Turkish reading 40.9% (95%: 37.2–44.7%) Accuracy 29.4% (95%: 27.5–31.3%) Reading 40.9% · grammar 24.5% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 65.5% · answered 45.0% · error rate 64.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.013$13.17 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.019 s median per request, 1 in flight | 40.9%37.2%–44.7% | 29.4% | 24.5% | 65.5%answered 45% | 5.8option score | $0.013$13.17from GPU time | 0.0191 per request1 in flight |
| 100 | OneJev1.6-0.8B‡ Open · 0.8BLatency 0.017 sCost No list price OneJev1.6-0.8B · OmniJev · decision model Parameters: 800M dense, as the maker states it (counted from the released files: 0.85B). Turkish reading 39.6% (95%: 36.0–43.4%) Accuracy 29.1% (95%: 27.3–31.1%) Reading 39.6% · grammar 24.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 43.4% · answered 65.7% · error rate 68.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.3 points (option score) Cost: No list price · 1 question per request Latency: 0.017 s median per request, 1 in flight | 39.6%36.0%–43.4% | 29.1% | 24.6% | 43.4%answered 66% | 9.3option score | No list price | 0.0171 per request1 in flight |
| 101 | KnowLine-0.8B-Gen1‡ Open · 0.8BLatency 0.033 sCost No list price KnowLine-0.8B-Gen1 · PelaAI · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish reading 38.3% (95%: 34.6–42.0%) Accuracy 27.1% (95%: 25.3–29.0%) Reading 38.3% · grammar 22.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 95.3% · error rate 72.3% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost: No list price · 1 question per request Latency: 0.033 s median per request, 1 in flight | 38.3%34.6%–42.0% | 27.1% | 22.4% | 6.4%answered 95% | 8.3option score | No list price | 0.0331 per request1 in flight |
| 102 | LFM3-1.7B reasoning off · chat‡ Open · 1.7BLatency 4.90 sCost $0.646$646 per 1,000per 1M LFM3-1.7B reasoning off · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish reading 37.2% (95%: 33.6–41.0%) Accuracy 28.3% (95%: 26.5–30.3%) Reading 37.2% · grammar 24.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 76.2% · answered 32.7% · error rate 62.4% Confidence: not measured Cost $0.646$646 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 4.90 s median per request, 32 in flight | 37.2%33.6%–41.0% | 28.3% | 24.6% | 76.2%answered 33% | Not measureda plain answer line, no confidence | $0.646$646at the maker's list price | 4.901 per request32 in flight |
| 103 | Strands Decider 2B‡ Open · 2.0BLatency 0.026 sCost $0.020$20.12 per 1,000per 1M Strands Decider 2B · Strands Agents · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.29B). Turkish reading 36.9% (95%: 33.3–40.6%) Accuracy 27.5% (95%: 25.7–29.4%) Reading 36.9% · grammar 23.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 93.2% · error rate 72.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 12.2 points (option score) Cost $0.020$20.12 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.026 s median per request, 1 in flight | 36.9%33.3%–40.6% | 27.5% | 23.5% | 9.2%answered 93% | 12.2option score | $0.020$20.12from GPU time | 0.0261 per request1 in flight |
| 104 | StartLux-Decision-v1.1-0.8B-Instant‡ Open · 0.8BLatency 0.028 sCost No list price StartLux-Decision-v1.1-0.8B-Instant · StartLux · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish reading 35.1% (95%: 31.5–38.8%) Accuracy 28.6% (95%: 26.7–30.5%) Reading 35.1% · grammar 25.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 12.4% · answered 90.9% · error rate 70.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.4 points (option score) Cost: No list price · 1 question per request Latency: 0.028 s median per request, 1 in flight | 35.1%31.5%–38.8% | 28.6% | 25.8% | 12.4%answered 91% | 4.4option score | No list price | 0.0281 per request1 in flight |
| 105 | Decider 1‡ Closed · EfficientLatency 0.382 sCost $0.015$14.76 per 1,000per 1M Decider 1 · meraGPT · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://meragpt.com/docs/models, read 2026-10-19) Turkish reading 34.6% (95%: 31.1–38.3%) Accuracy 27.1% (95%: 25.3–29.0%) Reading 34.6% · grammar 23.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 14.3 points (option score) Cost $0.015$14.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.382 s median per request, 1 in flight | 34.6%31.1%–38.3% | 27.1% | 23.9% | Not run | 14.3option score | $0.015$14.76at list prices | 0.3821 per request1 in flight |
| 106 | Qwen3.5-2B reasoning off · chat‡ Open · 2.0BLatency 8.84 sCost $0.268$268 per 1,000per 1M Qwen3.5-2B reasoning off · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish reading 34.3% (95%: 30.8–38.0%) Accuracy 25.4% (95%: 23.6–27.2%) Reading 34.3% · grammar 21.6% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 20.8% · answered 75.6% · error rate 72.3% Confidence: not measured Cost $0.268$268 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 8.84 s median per request, 32 in flight | 34.3%30.8%–38.0% | 25.4% | 21.6% | 20.8%answered 76% | Not measureda plain answer line, no confidence | $0.268$268from GPU time | 8.841 per request32 in flight |
| 107 | GLiClass v4 Edge‡ Open · 0.4BLatency 0.025 sCost No list price GLiClass v4 Edge · Knowledgator · decision model Base model: DeBERTa-v3-large Parameters: 400M dense, as the maker states it (counted from the released files: 0.44B). Turkish reading 25.9% (95%: 22.7–29.4%) Accuracy 23.3% (95%: 21.6–25.2%) Reading 25.9% · grammar 22.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.3% · answered 81.9% · error rate 77.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.5 points (option score) Cost: No list price · 1 question per request Latency: 0.025 s median per request, 1 in flight | 25.9%22.7%–29.4% | 23.3% | 22.2% | 22.3%answered 82% | 10.5option score | No list price | 0.0251 per request1 in flight |
| 108 | LFM2.5-8B-A1B default reasoning · chat‡ Open · 8.3B MoE · 1.5B activeLatency 39.2 sCost $1.050$1,050 per 1,000per 1M LFM2.5-8B-A1B default reasoning · chat · Liquid AI · chat model Mixture of experts: 8.3B total (counted from the released files: 8.47B), about 1.5B active per step. Size class uses the total (8.3B). Turkish reading 25.6% (95%: 22.4–29.1%) Accuracy 21.0% (95%: 19.3–22.7%) Reading 25.6% · grammar 19.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 56.0% · answered 42.4% · error rate 74.0% Confidence: not measured Cost $1.050$1,050 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 39.2 s median per request, 32 in flight | 25.6%22.4%–29.1% | 21.0% | 19.0% | 56.0%answered 42% | Not measureda plain answer line, no confidence | $1.050$1,050from GPU time | 39.21 per request32 in flight |
| 109 | GLiDE 0.4B‡ Open · 0.4BLatency 0.016 sCost No list price GLiDE 0.4B · Fastino · decision model Parameters: 400M dense, as the maker states it (counted from the released files: 0.40B). Turkish reading 24.5% (95%: 21.4–28.0%) Accuracy 22.2% (95%: 20.6–24.0%) Reading 24.5% · grammar 21.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 95.3% · answered 5.8% · error rate 78.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.0 points (option score) Cost: No list price · 1 question per request Latency: 0.016 s median per request, 1 in flight | 24.5%21.4%–28.0% | 22.2% | 21.3% | 95.3%answered 6% | 8.0option score | No list price | 0.0161 per request1 in flight |
| 110 | GLiNER2.5-multi-Decide‡ Open · 0.3BLatency 0.009 sCost $0.006$6.27 per 1,000per 1M GLiNER2.5-multi-Decide · Fastino · decision model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish reading 21.0% (95%: 18.1–24.3%) Accuracy 20.2% (95%: 18.6–21.9%) Reading 21.0% · grammar 19.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 94.8% · error rate 79.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 23.1 points (option score) Cost $0.006$6.27 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 21.0%18.1%–24.3% | 20.2% | 19.8% | 6.4%answered 95% | 23.1option score | $0.006$6.27from GPU time | 0.0091 per request1 in flight |
| 111 | Laya‡ Open · 0.4BLatency 0.011 sCost $0.009$9.00 per 1,000per 1M Laya · Convai Innovations · decision model Parameters: 421M dense, as the maker states it (counted from the released files: 0.42B). Turkish reading 20.9% (95%: 17.9–24.2%) Accuracy 19.0% (95%: 17.4–20.7%) Reading 20.9% · grammar 18.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 28.3% · answered 75.6% · error rate 81.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 13.2 points (option score) Cost $0.009$9.00 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.011 s median per request, 1 in flight | 20.9%17.9%–24.2% | 19.0% | 18.2% | 28.3%answered 76% | 13.2option score | $0.009$9.00from GPU time | 0.0111 per request1 in flighton its repeats |
| 112 | GLiNER2.5 Multi‡ Open · 0.3BLatency 0.032 sCost $0.006$5.88 per 1,000per 1M GLiNER2.5 Multi · Fastino · chat model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish reading 17.2% (95%: 14.5–20.3%) Accuracy 17.5% (95%: 15.9–19.1%) Reading 17.2% · grammar 17.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.3% · answered 94.7% · error rate 82.6% Confidence: calibration gap 35.6 points (option score) Cost $0.006$5.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.032 s median per request, 1 in flight | 17.2%14.5%–20.3% | 17.5% | 17.6% | 6.3%answered 95% | 35.6option score | $0.006$5.88at the maker's list price | 0.0321 per request1 in flight |
| 113 | Laya Multilingual‡ Open · 0.3BLatency 0.009 sCost $0.008$7.74 per 1,000per 1M Laya Multilingual · Convai Innovations · decision model Parameters: 322M dense, as the maker states it (counted from the released files: 0.32B). Turkish reading 17.1% (95%: 14.4–20.1%) Accuracy 17.9% (95%: 16.3–19.5%) Reading 17.1% · grammar 18.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.8% · answered 87.3% · error rate 83.8% Kept at 99% confidence: 0.0%, 1 of them wrong Confidence: calibration gap 27.7 points (option score) Cost $0.008$7.74 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 17.1%14.4%–20.1% | 17.9% | 18.2% | 14.8%answered 87% | 27.7option score | $0.008$7.74from GPU time | 0.0091 per request1 in flighton its repeats |
| 114 | MiniCPM5-2B reasoning off · chat‡ Open · 2.5BLatency 172.0 sCost $5.464$5,464 per 1,000per 1M MiniCPM5-2B reasoning off · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Turkish reading 6.6% (95%: 4.9–8.7%) Accuracy 5.6% (95%: 4.7–6.6%) Reading 6.6% · grammar 5.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 25.4% · answered 20.1% · error rate 75.3% Confidence: not measured Cost $5.464$5,464 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 172.0 s median per request, 32 in flight | 6.6%4.9%–8.7% | 5.6% | 5.2% | 25.4%answered 20% | Not measureda plain answer line, no confidence | $5.464$5,464from GPU time | 172.01 per request32 in flight |
Accuracy on reading: the share of the 656 Turkish reading questions answered right, with its 95% interval as the whisker; higher is better.
No row of this part matches these keys. The small open chat models ran in English only.
Rank: the row's place among the rows in view, by accuracy, 1, 2, 3 without repeats; rows with the same score keep the table's order, which claims no difference between them. No all-pairs test covers the Turkish reading part, so it has no tiers. A setup with web search is not ranked (Rank: –).
‡ Turkish not officially supported: the model's own card or maker docs do not list Turkish (read 16 November 2026; Ümit #5067); of the models here only Erk-14B and Metask-Jev-4B list it.
| Rank | Model | Accuracy on grammarof 1,542 questions, 95% interval | Turkishof 2,198 questions | Readingof 656 questions | Non-hallucination% “I don't know” of answers not right | Confidencecalibration gap, points | Costper 1,000 questionsper 1M questions | Latencymedian s per request |
|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon high · chat‡ Closed · FlagshipLatency 217.7 sCost $31.309$31,309 per 1,000per 1M Gemini 4 Carbon high · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish grammar 98.8% (95%: 98.2–99.3%) Accuracy 98.5% (95%: 97.8–98.9%) Reading 97.6% · grammar 98.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.9% · answered 99.7% · error rate 1.5% Confidence: calibration gap 0.4 points (written estimate) Cost $31.309$31,309 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 217.7 s median per request, 1 in flight | 98.8%98.2%–99.3% | 98.5% | 97.6% | 17.9%answered 100% | 0.4written estimate | $31.309$31,309subscription, priced as the API | 217.716 per request1 in flight |
| 2 | Gemini 3.8 Flash high · chat‡ Closed · WorkhorseLatency 53.7 sCost $2.419$2,419 per 1,000per 1M Gemini 3.8 Flash high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish grammar 98.5% (95%: 97.8–99.0%) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.1% · grammar 98.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.5% · answered 99.9% · error rate 1.9% Confidence: calibration gap 1.1 points (written estimate) Cost $2.419$2,419 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 53.7 s median per request, 4 in flight | 98.5%97.8%–99.0% | 98.1% | 97.1% | 4.5%answered 100% | 1.1written estimate | $2.419$2,419at list prices | 53.716 per request4 in flight |
| Gemini 3.8 Flash low · chat‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.19 sCost $0.269$269 per 1,000per 1M Gemini 3.8 Flash low · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish grammar 92.5% (95%: 91.1–93.7%) Accuracy 93.3% (95%: 92.1–94.2%) Reading 95.1% · grammar 92.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.9% · error rate 7.2% Confidence: calibration gap 3.5 points (written estimate) Cost $0.269$269 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 2.19 s median per request, 1 in flight | 92.5%91.1%–93.7% | 93.3% | 95.1% | 1.9%answered 100% | 3.5written estimate | $0.269$269at list prices | 2.1916 per request1 in flight | |
| Gemini 3.8 Flash low · chat · 1 per request‡ Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.72 sCost $0.963$963 per 1,000per 1M Gemini 3.8 Flash low · chat · 1 per request · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish grammar 97.5% (95%: 96.6–98.2%) Accuracy 97.1% (95%: 96.3–97.8%) Reading 96.2% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.2% · answered 99.8% · error rate 2.7% Confidence: calibration gap 0.6 points (written estimate) Cost $0.963$963 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.72 s median per request, 1 in flight | 97.5%96.6%–98.2% | 97.1% | 96.2% | 6.2%answered 100% | 0.6written estimate | $0.963$963at list prices | 2.721 per request1 in flight | |
| 3 | Gemini 4 Hydrogen high · chat‡ Closed · WorkhorseLatency 31.7 sCost $4.806$4,806 per 1,000per 1M Gemini 4 Hydrogen high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish grammar 98.4% (95%: 97.7–99.0%) Accuracy 98.1% (95%: 97.4–98.6%) Reading 97.3% · grammar 98.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 17.4% · answered 99.1% · error rate 1.7% Confidence: calibration gap 1.2 points (written estimate) Cost $4.806$4,806 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 31.7 s median per request, 1 in flight | 98.4%97.7%–99.0% | 98.1% | 97.3% | 17.4%answered 99% | 1.2written estimate | $4.806$4,806subscription, priced as the API | 31.716 per request1 in flight |
| 4 | GPT-6.1 Sol high · chat‡ Closed · WorkhorseLatency 24.7 sCost $1.622$1,622 per 1,000per 1M GPT-6.1 Sol high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 98.3% (95%: 97.5–98.8%) Accuracy 97.8% (95%: 97.1–98.3%) Reading 96.5% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 15.4% · answered 99.6% · error rate 2.0% Confidence: calibration gap 1.0 points (written estimate) Cost $1.622$1,622 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 24.7 s median per request, 1 in flight | 98.3%97.5%–98.8% | 97.8% | 96.5% | 15.4%answered 100% | 1.0written estimate | $1.622$1,622subscription, priced as the API | 24.716 per request1 in flight |
| GPT-6.1 Sol low · chat‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 13.3 sCost $1.130$1,130 per 1,000per 1M GPT-6.1 Sol low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 96.9% (95%: 95.9–97.6%) Accuracy 96.7% (95%: 95.8–97.4%) Reading 96.2% · grammar 96.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 99.7% · error rate 2.7% Confidence: calibration gap 1.5 points (written estimate) Cost $1.130$1,130 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.3 s median per request, 1 in flight | 96.9%95.9%–97.6% | 96.7% | 96.2% | 9.2%answered 100% | 1.5written estimate | $1.130$1,130subscription, priced as the API | 13.316 per request1 in flight | |
| GPT-6.1 Sol low · chat · 1 per request‡ Closed · Workhorseother setting of GPT-6.1 SolLatency 6.61 sCost $5.569$5,569 per 1,000per 1M GPT-6.1 Sol low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 97.0% (95%: 96.0–97.8%) Accuracy 96.7% (95%: 95.9–97.4%) Reading 96.0% · grammar 97.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.6% · answered 99.6% · error rate 2.8% Confidence: calibration gap 1.8 points (written estimate) Cost $5.569$5,569 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.61 s median per request, 1 in flight | 97.0%96.0%–97.8% | 96.7% | 96.0% | 11.6%answered 100% | 1.8written estimate | $5.569$5,569subscription, priced as the API | 6.611 per request1 in flight | |
| 5 | GPT-6 Astra high · chat‡ Closed · FlagshipLatency 21.7 sCost $6.066$6,066 per 1,000per 1M GPT-6 Astra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 98.3% (95%: 97.5–98.8%) Accuracy 98.0% (95%: 97.3–98.5%) Reading 97.1% · grammar 98.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.3% · answered 99.7% · error rate 1.9% Confidence: calibration gap 0.7 points (written estimate) Cost $6.066$6,066 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 21.7 s median per request, 1 in flight | 98.3%97.5%–98.8% | 98.0% | 97.1% | 14.3%answered 100% | 0.7written estimate | $6.066$6,066subscription, priced as the API | 21.716 per request1 in flight |
| GPT-6 Astra medium · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 15.3 sCost $5.447$5,447 per 1,000per 1M GPT-6 Astra medium · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 97.7% (95%: 96.9–98.4%) Accuracy 97.5% (95%: 96.8–98.1%) Reading 97.0% · grammar 97.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 12.7% · answered 99.7% · error rate 2.2% Confidence: calibration gap 1.0 points (written estimate) Cost $5.447$5,447 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 15.3 s median per request, 1 in flight | 97.7%96.9%–98.4% | 97.5% | 97.0% | 12.7%answered 100% | 1.0written estimate | $5.447$5,447subscription, priced as the API | 15.316 per request1 in flight | |
| GPT-6 Astra low · chat‡ Closed · Flagshipother setting of GPT-6 AstraLatency 13.0 sCost $5.378$5,378 per 1,000per 1M GPT-6 Astra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 97.9% (95%: 97.0–98.5%) Accuracy 97.6% (95%: 96.9–98.2%) Reading 97.0% · grammar 97.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 10.5% · answered 99.7% · error rate 2.3% Confidence: calibration gap 0.8 points (written estimate) Cost $5.378$5,378 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.0 s median per request, 1 in flight | 97.9%97.0%–98.5% | 97.6% | 97.0% | 10.5%answered 100% | 0.8written estimate | $5.378$5,378subscription, priced as the API | 13.016 per request1 in flighton a sample | |
| GPT-6 Astra low · chat · 1 per request‡ Closed · Flagshipother setting of GPT-6 AstraLatency 6.41 sCost $26.738$26,738 per 1,000per 1M GPT-6 Astra low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish grammar 97.5% (95%: 96.6–98.1%) Accuracy 97.4% (95%: 96.7–98.0%) Reading 97.3% · grammar 97.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 11.9% · answered 99.7% · error rate 2.4% Confidence: calibration gap 1.3 points (written estimate) Cost $26.738$26,738 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.41 s median per request, 1 in flight | 97.5%96.6%–98.1% | 97.4% | 97.3% | 11.9%answered 100% | 1.3written estimate | $26.738$26,738subscription, priced as the API | 6.411 per request1 in flight | |
| 6 | Claude Fable 5.2 max · chat‡ Closed · FlagshipLatency 81.1 sCost $53.381$53,381 per 1,000per 1M Claude Fable 5.2 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $8.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-11-11) Turkish grammar 98.3% (95%: 97.5–98.8%) Accuracy 98.1% (95%: 97.5–98.6%) Reading 97.7% · grammar 98.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know”: Queued (its run with the option has not started yet) Confidence: calibration gap 7.5 points (written estimate) Cost $53.381$53,381 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 81.1 s median per request, 1 in flight | 98.3%97.5%–98.8% | 98.1% | 97.7% | Queued | 7.5written estimate | $53.381$53,381subscription, priced as the API | 81.116 per request1 in flight |
| 7 | Claude Fable 5.1 high · chat‡ Closed · FlagshipLatency 30.3 sCost $17.042$17,042 per 1,000per 1M Claude Fable 5.1 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-28) Turkish grammar 96.4% (95%: 95.3–97.2%) Accuracy 96.1% (95%: 95.2–96.9%) Reading 95.6% · grammar 96.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.0% · answered 99.8% · error rate 3.6% Confidence: calibration gap 12.0 points (written estimate) Cost $17.042$17,042 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 30.3 s median per request, 1 in flight | 96.4%95.3%–97.2% | 96.1% | 95.6% | 6.0%answered 100% | 12.0written estimate | $17.042$17,042subscription, priced as the API | 30.316 per request1 in flight |
| 8 | Gemini 4 Argon low · chat‡ Closed · FlagshipLatency 5.46 sCost $1.786$1,786 per 1,000per 1M Gemini 4 Argon low · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-28) Turkish grammar 96.1% (95%: 95.0–97.0%) Accuracy 96.4% (95%: 95.5–97.1%) Reading 97.0% · grammar 96.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.0% · answered 99.7% · error rate 3.6% Confidence: calibration gap 2.2 points (written estimate) Cost $1.786$1,786 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.46 s median per request, 1 in flight | 96.1%95.0%–97.0% | 96.4% | 97.0% | 7.0%answered 100% | 2.2written estimate | $1.786$1,786at list prices | 5.4616 per request1 in flight |
| 9 | Claude Opus 5.5 high · chat‡ Closed · FlagshipLatency 22.1 sCost $6.509$6,509 per 1,000per 1M Claude Opus 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 94.7% (95%: 93.5–95.8%) Accuracy 95.0% (95%: 94.0–95.8%) Reading 95.6% · grammar 94.7% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 2.7% · answered 99.9% · error rate 4.9% Confidence: calibration gap 14.6 points (written estimate) Cost $6.509$6,509 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 22.1 s median per request, 1 in flight | 94.7%93.5%–95.8% | 95.0% | 95.6% | 2.7%answered 100% | 14.6written estimate | $6.509$6,509subscription, priced as the API | 22.116 per request1 in flight |
| Claude Opus 5.5 low · chat‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 13.4 sCost $4.931$4,931 per 1,000per 1M Claude Opus 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 88.6% (95%: 86.9–90.1%) Accuracy 90.4% (95%: 89.1–91.6%) Reading 94.8% · grammar 88.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.9% · error rate 7.7% Confidence: calibration gap 16.5 points (written estimate) Cost $4.931$4,931 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 13.4 s median per request, 1 in flight | 88.6%86.9%–90.1% | 90.4% | 94.8% | 1.8%answered 100% | 16.5written estimate | $4.931$4,931subscription, priced as the API | 13.416 per request1 in flighton a sample | |
| Claude Opus 5.5 low · chat · 1 per request‡ Closed · Flagshipother setting of Claude Opus 5.5Latency 3.71 sCost $11.610$11,610 per 1,000per 1M Claude Opus 5.5 low · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 92.3% (95%: 90.8–93.5%) Accuracy 93.3% (95%: 92.2–94.3%) Reading 95.7% · grammar 92.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 100.0% · error rate 6.2% Confidence: calibration gap 10.2 points (written estimate) Cost $11.610$11,610 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 3.71 s median per request, 1 in flight | 92.3%90.8%–93.5% | 93.3% | 95.7% | 0.7%answered 100% | 10.2written estimate | $11.610$11,610subscription, priced as the API | 3.711 per request1 in flight | |
| 10 | GPT-6.1 Terra high · chat‡ Closed · WorkhorseLatency 7.10 sCost $0.515$515 per 1,000per 1M GPT-6.1 Terra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish grammar 93.5% (95%: 92.1–94.6%) Accuracy 94.3% (95%: 93.2–95.2%) Reading 96.2% · grammar 93.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.3% · answered 99.6% · error rate 5.3% Confidence: calibration gap 2.8 points (written estimate) Cost $0.515$515 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.10 s median per request, 1 in flight | 93.5%92.1%–94.6% | 94.3% | 96.2% | 7.3%answered 100% | 2.8written estimate | $0.515$515subscription, priced as the API | 7.1016 per request1 in flight |
| GPT-6.1 Terra low · chat‡ Closed · Workhorseother setting of GPT-6.1 TerraLatency 4.24 sCost $0.426$426 per 1,000per 1M GPT-6.1 Terra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish grammar 88.7% (95%: 87.0–90.2%) Accuracy 90.9% (95%: 89.6–92.0%) Reading 95.9% · grammar 88.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.9% · answered 99.7% · error rate 9.1% Confidence: calibration gap 6.0 points (written estimate) Cost $0.426$426 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 4.24 s median per request, 1 in flight | 88.7%87.0%–90.2% | 90.9% | 95.9% | 2.9%answered 100% | 6.0written estimate | $0.426$426subscription, priced as the API | 4.2416 per request1 in flight | |
| 11 | Grok 4.7 high · chat‡ Closed · WorkhorseLatency 401.9 sCost $14.292$14,292 per 1,000per 1M Grok 4.7 high · chat · xAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://docs.x.ai/docs/models/grok-4.7, read 2026-10-28) Turkish grammar 91.8% (95%: 90.3–93.0%) Accuracy 92.7% (95%: 91.6–93.7%) Reading 95.0% · grammar 91.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.9% · answered 97.5% · error rate 6.1% Confidence: calibration gap 11.9 points (written estimate) Cost $14.292$14,292 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 401.9 s median per request, 4 in flight | 91.8%90.3%–93.0% | 92.7% | 95.0% | 29.9%answered 97% | 11.9written estimate | $14.292$14,292subscription, priced as the API | 401.916 per request4 in flight |
| 12 | DeepSeek-V4.1-Flash reasoning on · chat · first release‡ Open · 552B MoE · 16B activeLatency 365.1 sCost $1.299$1,299 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish grammar 91.4% (95%: 89.9–92.7%) Accuracy 92.0% (95%: 90.8–93.1%) Reading 93.4% · grammar 91.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.0 points (answer letter probability) Cost $1.299$1,299 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 365.1 s median per request, 8 in flight With its dated checkpoint deepseek-v4.1-flash-1017 under its other settings | 91.4%89.9%–92.7% | 92.0% | 93.4% | Not run | 8.0answer letter probability | $1.299$1,299at a host's list price for the same weights | 365.116 per request8 in flight |
| DeepSeek-V4.1-Flash reasoning on · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 486.9 sCost $1.278$1,278 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish grammar 90.8% (95%: 89.2–92.1%) Accuracy 91.8% (95%: 90.5–92.8%) Reading 94.1% · grammar 90.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.9% · error rate 7.0% Confidence: calibration gap 6.9 points (answer letter probability) Cost $1.278$1,278 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 486.9 s median per request, 8 in flight | 90.8%89.2%–92.1% | 91.8% | 94.1% | 1.3%answered 100% | 6.9answer letter probability | $1.278$1,278at the maker's list price | 486.916 per request8 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · first release‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 4.36 sCost $0.098$98.46 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · first release · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish grammar 56.6% (95%: 54.1–59.1%) Accuracy 65.9% (95%: 63.9–67.9%) Reading 87.8% · grammar 56.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.098$98.46 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 4.36 s median per request, 1 in flight | 56.6%54.1%–59.1% | 65.9% | 87.8% | Not run | 13.5answer letter probability | $0.098$98.46at a host's list price for the same weights | 4.3616 per request1 in flight | |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017‡ Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 5.08 sCost $0.098$98.47 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish grammar 56.5% (95%: 54.1–59.0%) Accuracy 66.0% (95%: 64.0–68.0%) Reading 88.3% · grammar 56.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 33.2% Confidence: calibration gap 16.4 points (answer letter probability) Cost $0.098$98.47 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 5.08 s median per request, 1 in flight | 56.5%54.1%–59.0% | 66.0% | 88.3% | 0.0%answered 100% | 16.4answer letter probability | $0.098$98.47at the maker's list price | 5.0816 per request1 in flight | |
| 13 | Qwen4-Max high · chat‡ Closed · WorkhorseLatency 91.5 sCost $2.533$2,533 per 1,000per 1M Qwen4-Max high · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://openrouter.ai/qwen/qwen4-max, read 2026-11-16) Turkish grammar 90.5% (95%: 88.9–91.8%) Accuracy 91.7% (95%: 90.5–92.8%) Reading 94.7% · grammar 90.5% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.2% · error rate 6.4% Confidence: calibration gap 7.8 points (written estimate) Cost $2.533$2,533 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 91.5 s median per request, 4 in flight | 90.5%88.9%–91.8% | 91.7% | 94.7% | 0.7%answered 99% | 7.8written estimate | $2.533$2,533at list prices | 91.516 per request4 in flight |
| 14 | Claude Sonnet 5.5 high · chat‡ Closed · WorkhorseLatency 20.5 sCost $3.654$3,654 per 1,000per 1M Claude Sonnet 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 85.0% (95%: 83.1–86.7%) Accuracy 85.0% (95%: 83.5–86.5%) Reading 85.2% · grammar 85.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.9% · error rate 12.8% Confidence: calibration gap 16.0 points (written estimate) Cost $3.654$3,654 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 20.5 s median per request, 1 in flight | 85.0%83.1%–86.7% | 85.0% | 85.2% | 0.7%answered 100% | 16.0written estimate | $3.654$3,654subscription, priced as the API | 20.516 per request1 in flighton a sample |
| Claude Sonnet 5.5 high · chat · 1 per request‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 4.64 sCost $7.748$7,748 per 1,000per 1M Claude Sonnet 5.5 high · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 87.0% (95%: 85.3–88.6%) Accuracy 89.2% (95%: 87.8–90.4%) Reading 94.2% · grammar 87.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.9% · error rate 11.1% Confidence: calibration gap 9.8 points (written estimate) Cost $7.748$7,748 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 4.64 s median per request, 1 in flight | 87.0%85.3%–88.6% | 89.2% | 94.2% | 0.8%answered 100% | 9.8written estimate | $7.748$7,748subscription, priced as the API | 4.641 per request1 in flight | |
| Claude Sonnet 5.5 low · chat‡ Closed · Workhorseother setting of Claude Sonnet 5.5Latency 3.87 sCost $1.896$1,896 per 1,000per 1M Claude Sonnet 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish grammar 52.2% (95%: 49.7–54.7%) Accuracy 61.4% (95%: 59.3–63.4%) Reading 82.9% · grammar 52.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.8% · error rate 35.1% Confidence: calibration gap 7.9 points (written estimate) Cost $1.896$1,896 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 3.87 s median per request, 1 in flight | 52.2%49.7%–54.7% | 61.4% | 82.9% | 0.6%answered 100% | 7.9written estimate | $1.896$1,896subscription, priced as the API | 3.8716 per request1 in flighton a sample | |
| 15 | Muse Ember max · chat‡ Closed · EfficientLatency 30.9 sCost $0.650$650 per 1,000per 1M Muse Ember max · chat · Meta · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/meta/muse-ember, read 2026-11-10) Turkish grammar 83.7% (95%: 81.7–85.4%) Accuracy 85.6% (95%: 84.1–87.0%) Reading 90.2% · grammar 83.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.1% · answered 97.5% · error rate 13.1% Confidence: calibration gap 6.2 points (written estimate) Cost $0.650$650 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 30.9 s median per request, 4 in flight | 83.7%81.7%–85.4% | 85.6% | 90.2% | 2.1%answered 98% | 6.2written estimate | $0.650$650at list prices | 30.916 per request4 in flight |
| 16 | Qwen3.5-397B-A17B high · chat‡ Open · 397B MoE · 17B activeLatency 251.3 sCost $2.411$2,411 per 1,000per 1M Qwen3.5-397B-A17B high · chat · Alibaba · chat model Mixture of experts: 397B total, about 17B active per step. Size class uses the total (397B). Turkish grammar 79.2% (95%: 77.1–81.1%) Accuracy 82.7% (95%: 81.1–84.2%) Reading 91.0% · grammar 79.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.5% · answered 99.7% · error rate 17.8% Confidence: calibration gap 6.2 points (written estimate) Cost $2.411$2,411 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 251.3 s median per request, 4 in flight | 79.2%77.1%–81.1% | 82.7% | 91.0% | 1.5%answered 100% | 6.2written estimate | $2.411$2,411at list prices | 251.316 per request4 in flight |
| 17 | Kimi K3.1 Pro low · chat‡ Open · 2.8T MoE · 104B activeLatency 44.9 sCost $2.048$2,048 per 1,000per 1M Kimi K3.1 Pro low · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish grammar 78.1% (95%: 75.9–80.1%) Accuracy 82.8% (95%: 81.1–84.3%) Reading 93.8% · grammar 78.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 2.0% · answered 99.7% · error rate 15.7% Confidence: calibration gap 12.6 points (written estimate) Cost $2.048$2,048 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 44.9 s median per request, 4 in flight | 78.1%75.9%–80.1% | 82.8% | 93.8% | 2.0%answered 100% | 12.6written estimate | $2.048$2,048at list prices | 44.916 per request4 in flight |
| Kimi K3.1 Pro no reasoning · chat‡ Open · 2.8T MoE · 104B activeother setting of Kimi K3.1 ProLatency 5.75 sCost $0.853$853 per 1,000per 1M Kimi K3.1 Pro no reasoning · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish grammar 64.3% (95%: 61.8–66.6%) Accuracy 72.7% (95%: 70.8–74.5%) Reading 92.5% · grammar 64.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.9% · error rate 28.3% Confidence: calibration gap 20.1 points (written estimate) Cost $0.853$853 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 5.75 s median per request, 4 in flight | 64.3%61.8%–66.6% | 72.7% | 92.5% | 0.5%answered 100% | 20.1written estimate | $0.853$853at list prices | 5.7516 per request4 in flight | |
| 18 | KnowLine-31B-Gen1.5‡ Open · 31BLatency 1.05 sCost No list price KnowLine-31B-Gen1.5 · PelaAI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 77.1% (95%: 74.9–79.1%) Accuracy 81.8% (95%: 80.1–83.4%) Reading 92.8% · grammar 77.1% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.2% · answered 100.0% · error rate 18.7% Kept at 99% confidence with no observed mistakes: 6.5% Confidence: calibration gap 3.4 points (option score) Cost: No list price · 1 question per request Latency: 1.05 s median per request, 1 in flight | 77.1%74.9%–79.1% | 81.8% | 92.8% | 0.2%answered 100% | 3.4option score | No list price | 1.051 per request1 in flight |
| 19 | Copernicus-31B‡ Open · 31BLatency 0.184 sCost $0.134$134 per 1,000per 1M Copernicus-31B · Poindexter Labs · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 77.0% (95%: 74.8–79.0%) Accuracy 81.8% (95%: 80.1–83.4%) Reading 93.1% · grammar 77.0% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.2% · answered 99.8% · error rate 18.3% Kept at 99% confidence with no observed mistakes: 4.9% Confidence: calibration gap 9.1 points (option score) Cost $0.134$134 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.184 s median per request, 1 in flight | 77.0%74.8%–79.0% | 81.8% | 93.1% | 1.2%answered 100% | 9.1option score | $0.134$134from GPU time | 0.1841 per request1 in flight |
| 20 | Inkling-Small high · chat · Jul 2026‡ Open · 276B MoE · 12B activeLatency 97.7 sCost $1.137$1,137 per 1,000per 1M Inkling-Small high · chat · Jul 2026 · Thinking Machines · chat model Mixture of experts: 276B total (counted from the released files: 265.96B), about 12B active per step. Size class uses the total (276B). Turkish grammar 76.9% (95%: 74.7–78.9%) Accuracy 80.0% (95%: 78.3–81.6%) Reading 87.2% · grammar 76.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.1% · answered 99.3% · error rate 21.4% Confidence: calibration gap 7.9 points (written estimate) Cost $1.137$1,137 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 97.7 s median per request, 4 in flight | 76.9%74.7%–78.9% | 80.0% | 87.2% | 3.1%answered 99% | 7.9written estimate | $1.137$1,137at list prices | 97.716 per request4 in flight |
| 21 | KnowLine-31B-Gen2‡ Open · 31BLatency 0.168 sCost $0.120$120 per 1,000per 1M KnowLine-31B-Gen2 · PelaAI · decision model Base model: Gemma 4 31B Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 76.0% (95%: 73.8–78.1%) Accuracy 81.2% (95%: 79.5–82.8%) Reading 93.4% · grammar 76.0% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 18.6% Kept at 99% confidence with no observed mistakes: 3.7% Confidence: calibration gap 5.6 points (option score) Cost $0.120$120 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.168 s median per request, 1 in flight | 76.0%73.8%–78.1% | 81.2% | 93.4% | 0.0%answered 100% | 5.6option score | $0.120$120from GPU time | 0.1681 per request1 in flight |
| 22 | Turion Jev 31B‡ Open · 31BLatency 0.405 sCost No list price Turion Jev 31B · Turion.AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 74.8% (95%: 72.5–76.9%) Accuracy 80.1% (95%: 78.4–81.7%) Reading 92.7% · grammar 74.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.3% · answered 99.3% · error rate 20.0% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 11.1 points (option score) Cost: No list price · 1 question per request Latency: 0.405 s median per request, 1 in flight | 74.8%72.5%–76.9% | 80.1% | 92.7% | 3.3%answered 99% | 11.1option score | No list price | 0.4051 per request1 in flight |
| 23 | GLM-5.3-Flash reasoning on · chat‡ Open · 320B MoE · 18B activeLatency 102.6 sCost $0.543$543 per 1,000per 1M GLM-5.3-Flash reasoning on · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish grammar 74.3% (95%: 72.1–76.4%) Accuracy 78.8% (95%: 77.0–80.5%) Reading 89.3% · grammar 74.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 91.9% · error rate 16.6% Confidence: calibration gap 4.5 points (written estimate) Cost $0.543$543 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 102.6 s median per request, 8 in flight | 74.3%72.1%–76.4% | 78.8% | 89.3% | 0.6%answered 92% | 4.5written estimate | $0.543$543at the maker's list price | 102.616 per request8 in flight |
| GLM-5.3-Flash reasoning off · chat‡ Open · 320B MoE · 18B activeother setting of GLM-5.3-FlashLatency 1.07 sCost $0.047$47.28 per 1,000per 1M GLM-5.3-Flash reasoning off · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish grammar 47.4% (95%: 44.9–49.9%) Accuracy 57.8% (95%: 55.7–59.8%) Reading 82.2% · grammar 47.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 41.0% Confidence: calibration gap 21.0 points (written estimate) Cost $0.047$47.28 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 1.07 s median per request, 1 in flight | 47.4%44.9%–49.9% | 57.8% | 82.2% | 0.3%answered 100% | 21.0written estimate | $0.047$47.28at the maker's list price | 1.0716 per request1 in flight | |
| 24 | REAX 1.3‡ Open · 57BLatency 0.250 sCost $0.185$185 per 1,000per 1M REAX 1.3 · REAX · decision model Two models run together, option probabilities averaged 50/50: 57.36B in all, all used on every question. Turkish grammar 71.8% (95%: 69.5–74.0%) Accuracy 77.9% (95%: 76.2–79.6%) Reading 92.4% · grammar 71.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 4.1 points (option score) Cost $0.185$185 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 71.8%69.5%–74.0% | 77.9% | 92.4% | 0.4%answered 100% | 4.1option score | $0.185$185from GPU time | 0.2501 per request1 in flight |
| 25 | Gemini Decision 1 20261111‡ Closed · EfficientLatency 0.227 sCost $0.037$37.19 per 1,000per 1M Gemini Decision 1 20261111 · Google · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/google/gemini-decision-1, read 2026-11-11) Turkish grammar 71.3% (95%: 69.0–73.5%) Accuracy 77.5% (95%: 75.7–79.2%) Reading 92.1% · grammar 71.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.5% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 14.5% Confidence: calibration gap 2.0 points (option score) Cost $0.037$37.19 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.227 s median per request, 1 in flight | 71.3%69.0%–73.5% | 77.5% | 92.1% | 1.9%answered 100% | 2.0option score | $0.037$37.19at list prices | 0.2271 per request1 in flight |
| 26 | Karpo Decision 31B‡ Open · 31BLatency 0.159 sCost $0.105$105 per 1,000per 1M Karpo Decision 31B · MachinePulse · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 71.1% (95%: 68.8–73.3%) Accuracy 77.8% (95%: 76.0–79.5%) Reading 93.4% · grammar 71.1% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.9% · error rate 21.6% Kept at 99% confidence with no observed mistakes: 23.0% Confidence: calibration gap 2.3 points (option score) Cost $0.105$105 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.159 s median per request, 1 in flight | 71.1%68.8%–73.3% | 77.8% | 93.4% | 0.6%answered 100% | 2.3option score | $0.105$105from GPU time | 0.1591 per request1 in flight |
| 27 | d1-ultra‡ Closed · EfficientLatency 0.223 sCost $0.014$14.34 per 1,000per 1M d1-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-ultra, read 2026-10-16) Turkish grammar 70.7% (95%: 68.4–72.9%) Accuracy 77.1% (95%: 75.3–78.8%) Reading 92.1% · grammar 70.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 75.1% · answered 71.9% · error rate 13.0% Kept at 99% confidence with no observed mistakes: 4.1% Confidence: calibration gap 3.6 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.223 s median per request, 1 in flight | 70.7%68.4%–72.9% | 77.1% | 92.1% | 75.1%answered 72% | 3.6option score | $0.014$14.34at list prices | 0.2231 per request1 in flight |
| 28 | GLM-5.3 low · chat‡ Open · WorkhorseLatency 17.1 sCost $0.814$814 per 1,000per 1M GLM-5.3 low · chat · Z.ai · chat model Parameters: not stated by the maker (a mixture of experts); counted from the released files: 753.33B. Class: Workhorse, by its standard list price: $1.40 per 1M input tokens (https://openrouter.ai/z-ai/glm-5.3, read 2026-10-28) Turkish grammar 68.9% (95%: 66.6–71.2%) Accuracy 74.5% (95%: 72.7–76.3%) Reading 87.7% · grammar 68.9% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 1.1% · answered 99.0% · error rate 23.9% Confidence: calibration gap 2.6 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 17.1 s median per request, 4 in flight | 68.9%66.6%–71.2% | 74.5% | 87.7% | 1.1%answered 99% | 2.6written estimate | $0.814$814at list prices | 17.116 per request4 in flight |
| 29 | Qwen3.8-27B low · chat‡ Open · 27BLatency 648.4 sCost $3.095$3,095 per 1,000per 1M Qwen3.8-27B low · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish grammar 68.4% (95%: 66.1–70.7%) Accuracy 74.7% (95%: 72.8–76.4%) Reading 89.3% · grammar 68.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.9% · answered 97.9% · error rate 22.9% Confidence: calibration gap 23.7 points (answer letter probability) Cost $3.095$3,095 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 648.4 s median per request, 4 in flight | 68.4%66.1%–70.7% | 74.7% | 89.3% | 5.9%answered 98% | 23.7answer letter probability | $3.095$3,095at the maker's list price | 648.416 per request4 in flight |
| Qwen3.8-27B yes/no scoring‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.280 sCost $0.704$704 per 1,000per 1M Qwen3.8-27B yes/no scoring · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish grammar 39.8% (95%: 37.3–42.2%) Accuracy 51.9% (95%: 49.8–54.0%) Reading 80.5% · grammar 39.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Confidence: calibration gap 8.9 points (option score) Cost $0.704$704 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.280 s median per request, 1 in flight | 39.8%37.3%–42.2% | 51.9% | 80.5% | Not run | 8.9option score | $0.704$704at the maker's list price | 0.2801 per request1 in flighton its repeats | |
| Qwen3.8-27B stated confidence · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 0.433 sCost $0.234$234 per 1,000per 1M Qwen3.8-27B stated confidence · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish grammar 41.8% (95%: 39.3–44.2%) Accuracy 54.2% (95%: 52.1–56.3%) Reading 83.4% · grammar 41.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.6% · answered 99.7% · error rate 45.6% Confidence: calibration gap 40.8 points (written estimate) Cost $0.234$234 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.433 s median per request, 1 in flight | 41.8%39.3%–44.2% | 54.2% | 83.4% | 0.6%answered 100% | 40.8written estimate | $0.234$234at the maker's list price | 0.4331 per request1 in flighton its repeats | |
| Qwen3.8-27B vote share · reasoning off · chat‡ Open · 27Bother setting of Qwen3.8-27BLatency 1.62 sCost $0.825$825 per 1,000per 1M Qwen3.8-27B vote share · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish grammar 41.7% (95%: 39.3–44.2%) Accuracy 54.3% (95%: 52.2–56.4%) Reading 84.0% · grammar 41.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.3% · answered 99.4% · error rate 45.6% Confidence: calibration gap 22.5 points (agreement across ten answers) Cost $0.825$825 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 1.62 s median per request, 1 in flight | 41.7%39.3%–44.2% | 54.3% | 84.0% | 1.3%answered 99% | 22.5agreement across ten answers | $0.825$825at the maker's list price | 1.621 per request1 in flight | |
| 30 | Logits Sigma-1 Duo‡ Open · 57BLatency 0.250 sCost $0.181$181 per 1,000per 1M Logits Sigma-1 Duo · Logits · decision model Parameters: 57.36B dense, as the maker states it. Turkish grammar 67.8% (95%: 65.5–70.1%) Accuracy 75.0% (95%: 73.2–76.8%) Reading 91.9% · grammar 67.8% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 24.7% Kept at 99% confidence with no observed mistakes: 1.9% Confidence: calibration gap 5.7 points (option score) Cost $0.181$181 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.250 s median per request, 1 in flight | 67.8%65.5%–70.1% | 75.0% | 91.9% | 0.4%answered 100% | 5.7option score | $0.181$181from GPU time | 0.2501 per request1 in flight |
| 31 | Sieve 31B v3‡ Open · 31BLatency 1.00 sCost No list price Sieve 31B v3 · Sthānika AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 66.7% (95%: 64.3–69.0%) Accuracy 74.2% (95%: 72.3–75.9%) Reading 91.8% · grammar 66.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 25.2% Kept at 99% confidence: 15.0%, 1 of them wrong Confidence: calibration gap 2.9 points (option score) Cost: No list price · 1 question per request Latency: 1.00 s median per request, 1 in flight | 66.7%64.3%–69.0% | 74.2% | 91.8% | 0.0%answered 100% | 2.9option score | No list price | 1.001 per request1 in flight |
| 32 | MiMo-V2.6-Flash high · chat‡ Open · 309B MoE · 15B activeLatency 21.4 sCost $2.219$2,219 per 1,000per 1M MiMo-V2.6-Flash high · chat · Xiaomi · chat model Mixture of experts: 309B total, about 15B active per step. Size class uses the total (309B). Turkish grammar 66.0% (95%: 63.6–68.3%) Accuracy 72.1% (95%: 70.2–73.9%) Reading 86.6% · grammar 66.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 3.5% · answered 96.7% · error rate 26.2% Confidence: calibration gap 7.1 points (written estimate) Cost $2.219$2,219 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 21.4 s median per request, 32 in flight | 66.0%63.6%–68.3% | 72.1% | 86.6% | 3.5%answered 97% | 7.1written estimate | $2.219$2,219at list prices | 21.41 per request32 in flight |
| 33 | Gemma 4 31B reasoning off · chat‡ Open · 31BLatency 3.97 sCost $0.048$48.22 per 1,000per 1M Gemma 4 31B reasoning off · chat · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 65.6% (95%: 63.2–67.9%) Accuracy 73.1% (95%: 71.2–74.9%) Reading 90.9% · grammar 65.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.1% · answered 98.9% · error rate 26.1% Confidence: calibration gap 24.2 points (answer letter probability) Cost $0.048$48.22 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 3.97 s median per request, 1 in flight | 65.6%63.2%–67.9% | 73.1% | 90.9% | 4.1%answered 99% | 24.2answer letter probability | $0.048$48.22at a host's list price for the same weights | 3.9716 per request1 in flight |
| Gemma 4 31B reasoning off · chat · 1 per request‡ Open · 31Bother setting of Gemma 4 31BLatency 0.298 sCost $0.068$68.19 per 1,000per 1M Gemma 4 31B reasoning off · chat · 1 per request · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish grammar 57.3% (95%: 54.8–59.8%) Accuracy 66.8% (95%: 64.8–68.7%) Reading 89.0% · grammar 57.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.4% · error rate 33.5% Confidence: calibration gap 31.2 points (answer letter probability) Cost $0.068$68.19 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.298 s median per request, 1 in flight | 57.3%54.8%–59.8% | 66.8% | 89.0% | 1.9%answered 99% | 31.2answer letter probability | $0.068$68.19at a host's list price for the same weights | 0.2981 per request1 in flight | |
| 34 | GLiDE 27B‡ Open · 27BLatency 0.542 sCost No list price GLiDE 27B · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 62.7% (95%: 60.3–65.1%) Accuracy 71.1% (95%: 69.2–73.0%) Reading 90.9% · grammar 62.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.8% · answered 86.8% · error rate 23.1% Kept at 99% confidence with no observed mistakes: 11.2% Confidence: calibration gap 2.3 points (option score) Cost: No list price · 1 question per request Latency: 0.542 s median per request, 1 in flight | 62.7%60.3%–65.1% | 71.1% | 90.9% | 39.8%answered 87% | 2.3option score | No list price | 0.5421 per request1 in flight |
| 35 | GPT-6 Luna low · chat‡ Closed · EfficientLatency 7.71 sCost $0.048$48.01 per 1,000per 1M GPT-6 Luna low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish grammar 60.2% (95%: 57.7–62.6%) Accuracy 68.3% (95%: 66.3–70.2%) Reading 87.3% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.0% · answered 99.7% · error rate 33.2% Confidence: calibration gap 25.7 points (written estimate) Cost $0.048$48.01 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.71 s median per request, 1 in flight | 60.2%57.7%–62.6% | 68.3% | 87.3% | 1.0%answered 100% | 25.7written estimate | $0.048$48.01subscription, priced as the API | 7.7116 per request1 in flight |
| – | Perplexity Decider 27B v1.2 API · with web‡ Open · 27BLatency 0.482 sCost ≤ $1.034≤ $1,034 per 1,000per 1M Perplexity Decider 27B v1.2 API · with web · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 60.2% (95%: 57.7–62.6%) Accuracy 68.7% (95%: 66.8–70.6%) Reading 88.9% · grammar 60.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 99.0% · error rate 32.3% Kept at 99% confidence: 16.0%, 1 of them wrong Confidence: calibration gap 3.6 points (option score) Cost ≤ $1.034≤ $1,034 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.482 s median per request, 1 in flight | 60.2%57.7%–62.6% | 68.7% | 88.9% | 3.2%answered 99% | 3.6option score | ≤ $1.034≤ $1,034at list prices | 0.4821 per request1 in flight |
| 36 | Perplexity Decider 27B v1.3‡ Open · 27BLatency 0.110 sCost $0.081$81.18 per 1,000per 1M Perplexity Decider 27B v1.3 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 59.7% (95%: 57.2–62.1%) Accuracy 68.4% (95%: 66.5–70.3%) Reading 89.0% · grammar 59.7% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 0.7% · answered 99.8% · error rate 32.4% Kept at 99% confidence with no observed mistakes: 12.6% Confidence: calibration gap 5.0 points (option score) Cost $0.081$81.18 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.110 s median per request, 1 in flight | 59.7%57.2%–62.1% | 68.4% | 89.0% | 0.7%answered 100% | 5.0option score | $0.081$81.18from GPU time | 0.1101 per request1 in flight |
| 37 | GLiDE v1.2‡ Open · 27BLatency 0.413 sCost $0.014$14.37 per 1,000per 1M GLiDE v1.2 · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 59.5% (95%: 57.0–61.9%) Accuracy 68.5% (95%: 66.5–70.4%) Reading 89.6% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 39.5% · answered 85.4% · error rate 26.2% Kept at 99% confidence with no observed mistakes: 10.9% Confidence: calibration gap 2.9 points (option score) Cost $0.014$14.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.413 s median per request, 1 in flight | 59.5%57.0%–61.9% | 68.5% | 89.6% | 39.5%answered 85% | 2.9option score | $0.014$14.37at the maker's list price | 0.4131 per request1 in flight |
| 38 | MiniMax-M3.1-Flash high · chat‡ Open · 428B MoE · 23B activeLatency 112.6 sCost $0.814$814 per 1,000per 1M MiniMax-M3.1-Flash high · chat · MiniMax · chat model Mixture of experts: 428B total, about 23B active per step. Size class uses the total (428B). Turkish grammar 59.5% (95%: 57.0–61.9%) Accuracy 66.5% (95%: 64.5–68.4%) Reading 82.9% · grammar 59.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 5.5% · answered 97.6% · error rate 32.8% Confidence: calibration gap 21.1 points (written estimate) Cost $0.814$814 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 112.6 s median per request, 4 in flight | 59.5%57.0%–61.9% | 66.5% | 82.9% | 5.5%answered 98% | 21.1written estimate | $0.814$814at list prices | 112.616 per request4 in flight |
| 39 | GLiDE API‡ Closed · EfficientLatency 2.09 sCost $0.160$160 per 1,000per 1M GLiDE API · Fastino · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.fastino.ai/pricing, read 2026-10-22) Turkish grammar 59.4% (95%: 56.9–61.8%) Accuracy 67.9% (95%: 65.9–69.8%) Reading 87.8% · grammar 59.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 59.4% · answered 76.3% · error rate 21.2% Kept at 99% confidence with no observed mistakes: 5.5% Confidence: calibration gap 4.5 points (option score) Cost $0.160$160 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.09 s median per request, 4 in flight | 59.4%56.9%–61.8% | 67.9% | 87.8% | 59.4%answered 76% | 4.5option score | $0.160$160at list prices | 2.091 per request4 in flight |
| 40 | d1.2-ultra‡ Closed · EfficientLatency 0.220 sCost $0.014$14.34 per 1,000per 1M d1.2-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish grammar 58.9% (95%: 56.4–61.3%) Accuracy 68.1% (95%: 66.1–70.0%) Reading 89.8% · grammar 58.9% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 87.3% · answered 54.7% · error rate 12.1% Kept at 99% confidence with no observed mistakes: 0.5% Confidence: calibration gap 3.4 points (option score) Cost $0.014$14.34 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.220 s median per request, 1 in flight | 58.9%56.4%–61.3% | 68.1% | 89.8% | 87.3%answered 55% | 3.4option score | $0.014$14.34at list prices | 0.2201 per request1 in flight |
| 41 | d3‡ Open · 26BLatency 0.092 sCost $0.070$70.21 per 1,000per 1M d3 · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 58.7% (95%: 56.2–61.1%) Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.2% · grammar 58.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 32.5% Kept at 99% confidence with no observed mistakes: 11.4% Confidence: calibration gap 2.6 points (option score) Cost $0.070$70.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.092 s median per request, 1 in flight | 58.7%56.2%–61.1% | 67.8% | 89.2% | 0.0%answered 100% | 2.6option score | $0.070$70.21from GPU time | 0.0921 per request1 in flight |
| 42 | d3-ultra‡ Open · 26BLatency 0.333 sCost $0.216$216 per 1,000per 1M d3-ultra · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 58.4% (95%: 56.0–60.9%) Accuracy 67.8% (95%: 65.8–69.7%) Reading 89.8% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.5% · error rate 32.2% Kept at 99% confidence: 17.5%, 1 of them wrong Confidence: calibration gap 12.5 points (option score) Cost $0.216$216 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.333 s median per request, 1 in flight | 58.4%56.0%–60.9% | 67.8% | 89.8% | 1.4%answered 100% | 12.5option score | $0.216$216from GPU time | 0.3331 per request1 in flight |
| 43 | Drex Max‡ Open · 27BLatency 0.082 sCost $0.023$23.29 per 1,000per 1M Drex Max · Nace.AI · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 58.4% (95%: 56.0–60.9%) Accuracy 67.7% (95%: 65.8–69.7%) Reading 89.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.4% Kept at 99% confidence: 20.2%, 1 of them wrong Confidence: calibration gap 6.3 points (option score) Cost $0.023$23.29 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.082 s median per request, 1 in flight | 58.4%56.0%–60.9% | 67.7% | 89.6% | 0.3%answered 100% | 6.3option score | $0.023$23.29at the maker's list price | 0.0821 per request1 in flight |
| 44 | Drex Max v1.1‡ Closed · Efficient (provisional)Latency 0.313 sCost $0.023$23.11 per 1,000per 1M Drex Max v1.1 · Nace.AI · decision model Parameters: not disclosed by maker Class: Efficient (provisional), by its introductory price, the only one published: $0.03 per 1M input tokens (https://nace.ai/developers, read 2026-11-10) Turkish grammar 58.4% (95%: 55.9–60.8%) Accuracy 67.4% (95%: 65.4–69.3%) Reading 88.6% · grammar 58.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.2% · answered 85.4% · error rate 27.5% Kept at 99% confidence with no observed mistakes: 10.0% Confidence: calibration gap 2.9 points (option score) Cost $0.023$23.11 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 58.4%55.9%–60.8% | 67.4% | 88.6% | 38.2%answered 85% | 2.9option score | $0.023$23.11at list prices | 0.3131 per request1 in flight |
| 45 | Perplexity Decider 27B v1.2 DGX Spark‡ Open · 27BLatency 0.580 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.2 DGX Spark · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 57.7% (95%: 55.2–60.1%) Accuracy 67.1% (95%: 65.1–69.0%) Reading 89.2% · grammar 57.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.9% · error rate 32.7% Kept at 99% confidence with no observed mistakes: 16.4% Confidence: calibration gap 4.4 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.580 s median per request, 1 in flight | 57.7%55.2%–60.1% | 67.1% | 89.2% | 0.3%answered 100% | 4.4option score | $0.008$7.76at the maker's list price | 0.5801 per request1 in flight |
| Perplexity Decider 27B v1.2 API‡ Open · 27Bother setting of Perplexity Decider 27B v1.2Latency 0.254 sCost $0.008$7.70 per 1,000per 1M Perplexity Decider 27B v1.2 API · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 56.7% (95%: 54.2–59.1%) Accuracy 66.3% (95%: 64.3–68.3%) Reading 89.0% · grammar 56.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.9% · error rate 33.1% Kept at 99% confidence with no observed mistakes: 14.9% Confidence: calibration gap 4.2 points (option score) Cost $0.008$7.70 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.254 s median per request, 1 in flight | 56.7%54.2%–59.1% | 66.3% | 89.0% | 0.4%answered 100% | 4.2option score | $0.008$7.70at list prices | 0.2541 per request1 in flight | |
| 46 | Perplexity Decider 27B v1.1‡ Open · 27BLatency 0.233 sCost $0.008$7.76 per 1,000per 1M Perplexity Decider 27B v1.1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 57.5% (95%: 55.0–59.9%) Accuracy 66.9% (95%: 64.9–68.8%) Reading 89.0% · grammar 57.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 15.9% Confidence: calibration gap 3.8 points (option score) Cost $0.008$7.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.233 s median per request, 1 in flight | 57.5%55.0%–59.9% | 66.9% | 89.0% | Not run | 3.8option score | $0.008$7.76at list prices | 0.2331 per request1 in flight |
| 47 | Mammoth-1‡ Open · 27BLatency 0.169 sCost $0.040$39.62 per 1,000per 1M Mammoth-1 · Base Labs · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish grammar 57.2% (95%: 54.7–59.6%) Accuracy 67.1% (95%: 65.1–69.0%) Reading 90.4% · grammar 57.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 68.7% · answered 68.6% · error rate 20.8% Kept at 99% confidence with no observed mistakes: 13.0% Confidence: calibration gap 5.6 points (option score) Cost $0.040$39.62 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.169 s median per request, 1 in flight | 57.2%54.7%–59.6% | 67.1% | 90.4% | 68.7%answered 69% | 5.6option score | $0.040$39.62at the maker's list price | 0.1691 per request1 in flight |
| 48 | SubQ Snap Preview‡ Closed · EfficientLatency 0.389 sCost $0.019$19.41 per 1,000per 1M SubQ Snap Preview · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.05 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-10-13) Turkish grammar 55.7% (95%: 53.2–58.2%) Accuracy 65.2% (95%: 63.2–67.2%) Reading 87.7% · grammar 55.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.9 points (option score) Cost $0.019$19.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.389 s median per request, 1 in flight | 55.7%53.2%–58.2% | 65.2% | 87.7% | Not run | 4.9option score | $0.019$19.41at list prices | 0.3891 per request1 in flight |
| 49 | GPT-6 Luna Decisions API‡ Closed · EfficientLatency 0.191 sCost $0.040$40.44 per 1,000per 1M GPT-6 Luna Decisions API · OpenAI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish grammar 54.3% (95%: 51.8–56.8%) Accuracy 63.6% (95%: 61.6–65.6%) Reading 85.5% · grammar 54.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 16.7%, 3 of them wrong Confidence: calibration gap 2.4 points (option score) Cost $0.040$40.44 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.191 s median per request, 4 in flight | 54.3%51.8%–56.8% | 63.6% | 85.5% | Not run | 2.4option score | $0.040$40.44at list prices | 0.1911 per request4 in flight |
| 50 | SubQ Snap 1.1‡ Closed · EfficientLatency 0.378 sCost $0.015$15.41 per 1,000per 1M SubQ Snap 1.1 · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-11-01) Turkish grammar 54.0% (95%: 51.5–56.5%) Accuracy 63.9% (95%: 61.8–65.9%) Reading 87.0% · grammar 54.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 47.1% · answered 80.1% · error rate 27.8% Kept at 99% confidence: 13.6%, 1 of them wrong Confidence: calibration gap 11.2 points (option score) Cost $0.015$15.41 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.378 s median per request, 1 in flight | 54.0%51.5%–56.5% | 63.9% | 87.0% | 47.1%answered 80% | 11.2option score | $0.015$15.41at list prices | 0.3781 per request1 in flight |
| 51 | Qwen3.8 Flash no reasoning · chat‡ Closed · EfficientLatency 10.0 sCost $0.047$47.13 per 1,000per 1M Qwen3.8 Flash no reasoning · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.15 per 1M input tokens (https://openrouter.ai/qwen/qwen3.8-flash, read 2026-11-11) Turkish grammar 52.8% (95%: 50.3–55.3%) Accuracy 63.1% (95%: 61.0–65.1%) Reading 87.2% · grammar 52.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 98.3% · error rate 35.4% Confidence: calibration gap 28.7 points (written estimate) Cost $0.047$47.13 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 10.0 s median per request, 4 in flight | 52.8%50.3%–55.3% | 63.1% | 87.2% | 4.7%answered 98% | 28.7written estimate | $0.047$47.13at list prices | 10.016 per request4 in flight |
| 52 | KnowLine-12B-Gen1‡ Open · 12BLatency 0.071 sCost $0.052$52.28 per 1,000per 1M KnowLine-12B-Gen1 · PelaAI · decision model Base model: Gemma 4 12B Unified Parameters: 12B dense, as the maker states it (counted from the released files: 11.96B). Turkish grammar 52.5% (95%: 50.0–54.9%) Accuracy 62.5% (95%: 60.4–64.5%) Reading 86.0% · grammar 52.5% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 36.6% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.2 points (option score) Cost $0.052$52.28 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.071 s median per request, 1 in flight | 52.5%50.0%–54.9% | 62.5% | 86.0% | 0.1%answered 100% | 6.2option score | $0.052$52.28from GPU time | 0.0711 per request1 in flight |
| 53 | d1.1‡ Closed · EfficientLatency 0.195 sCost $0.007$6.83 per 1,000per 1M d1.1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-24) Turkish grammar 51.7% (95%: 49.2–54.2%) Accuracy 61.5% (95%: 59.5–63.5%) Reading 84.6% · grammar 51.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 4.0% Confidence: calibration gap 2.0 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.195 s median per request, 1 in flight | 51.7%49.2%–54.2% | 61.5% | 84.6% | Not run | 2.0option score | $0.007$6.83at list prices | 0.1951 per request1 in flight |
| 54 | d1.2‡ Closed · EfficientLatency 0.213 sCost $0.007$6.83 per 1,000per 1M d1.2 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish grammar 51.4% (95%: 48.9–53.9%) Accuracy 61.7% (95%: 59.7–63.7%) Reading 86.0% · grammar 51.4% ‡ Turkish not officially supported: it names no language (source, read 16 November 2026) Non-hallucination with “I don't know” 79.4% · answered 55.7% · error rate 20.7% Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 3.3 points (option score) Cost $0.007$6.83 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.213 s median per request, 1 in flight | 51.4%48.9%–53.9% | 61.7% | 86.0% | 79.4%answered 56% | 3.3option score | $0.007$6.83at list prices | 0.2131 per request1 in flight |
| 55 | Qwen3.5-4B reasoning on · chat‡ Open · 4.0BLatency 140.3 sCost $2.998$2,998 per 1,000per 1M Qwen3.5-4B reasoning on · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish grammar 50.3% (95%: 47.8–52.8%) Accuracy 58.9% (95%: 56.8–61.0%) Reading 79.1% · grammar 50.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.3% · answered 98.4% · error rate 40.7% Confidence: not measured Cost $2.998$2,998 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 140.3 s median per request, 32 in flight | 50.3%47.8%–52.8% | 58.9% | 79.1% | 3.3%answered 98% | Not measureda plain answer line, no confidence | $2.998$2,998from GPU time | 140.31 per request32 in flight |
| 56 | celeris-1-decision‡ Closed · EfficientLatency 0.401 sCost $0.015$15.28 per 1,000per 1M celeris-1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-10-10) Turkish grammar 49.9% (95%: 47.4–52.4%) Accuracy 60.5% (95%: 58.4–62.5%) Reading 85.2% · grammar 49.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.3 points (option score) Cost $0.015$15.28 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.401 s median per request, 1 in flight | 49.9%47.4%–52.4% | 60.5% | 85.2% | Not run | 4.3option score | $0.015$15.28at list prices | 0.4011 per request1 in flight |
| 57 | Perplexity Decider 27B v1‡ Open · 27BLatency 0.216 sCost $0.016$15.54 per 1,000per 1M Perplexity Decider 27B v1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish grammar 49.5% (95%: 47.0–52.0%) Accuracy 60.4% (95%: 58.3–62.4%) Reading 86.0% · grammar 49.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.0% Confidence: calibration gap 5.0 points (option score) Cost $0.016$15.54 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.216 s median per request, 1 in flight | 49.5%47.0%–52.0% | 60.4% | 86.0% | Not run | 5.0option score | $0.016$15.54at list prices | 0.2161 per request1 in flighton its repeats |
| 58 | Gemma 4 E4B default reasoning · chat‡ Open · 8.0BLatency 13.4 sCost $0.307$307 per 1,000per 1M Gemma 4 E4B default reasoning · chat · Google · chat model Parameters: 8B dense · 4.5B effective, as the maker states it (counted from the released files: 8.00B). Turkish grammar 49.0% (95%: 46.5–51.5%) Accuracy 57.6% (95%: 55.5–59.6%) Reading 77.6% · grammar 49.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.7% · answered 86.0% · error rate 38.4% Confidence: not measured Cost $0.307$307 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 13.4 s median per request, 32 in flight | 49.0%46.5%–51.5% | 57.6% | 77.6% | 29.7%answered 86% | Not measureda plain answer line, no confidence | $0.307$307from GPU time | 13.41 per request32 in flight |
| 59 | OneJev-27B‡ Open · 27BLatency 0.125 sCost $0.099$99.09 per 1,000per 1M OneJev-27B · OmniJev · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish grammar 48.7% (95%: 46.2–51.2%) Accuracy 59.6% (95%: 57.5–61.6%) Reading 85.2% · grammar 48.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 40.6% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 5.5 points (option score) Cost $0.099$99.09 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.125 s median per request, 1 in flight | 48.7%46.2%–51.2% | 59.6% | 85.2% | 0.1%answered 100% | 5.5option score | $0.099$99.09from GPU time | 0.1251 per request1 in flight |
| 60 | Jev 1.14‡ Closed · EfficientLatency 0.211 sCost $0.029$28.84 per 1,000per 1M Jev 1.14 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://docs.typesafe.ai/models, read 2026-10-16) Turkish grammar 48.2% (95%: 45.7–50.7%) Accuracy 59.2% (95%: 57.1–61.2%) Reading 85.1% · grammar 48.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 11.8%, 1 of them wrong Confidence: calibration gap 2.5 points (option score) Cost $0.029$28.84 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.211 s median per request, 1 in flight | 48.2%45.7%–50.7% | 59.2% | 85.1% | Not run | 2.5option score | $0.029$28.84at list prices | 0.2111 per request1 in flight |
| 61 | Mercury Decide‡ Closed · EfficientLatency 0.313 sCost $0.008$7.71 per 1,000per 1M Mercury Decide · Inception · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://www.inceptionlabs.ai/, read 2026-11-10) Turkish grammar 47.1% (95%: 44.7–49.6%) Accuracy 58.3% (95%: 56.2–60.3%) Reading 84.5% · grammar 47.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 20.5%, 13 of them wrong Confidence: calibration gap 15.5 points (option score) Cost $0.008$7.71 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.313 s median per request, 1 in flight | 47.1%44.7%–49.6% | 58.3% | 84.5% | Not run | 15.5option score | $0.008$7.71at list prices | 0.3131 per request1 in flight |
| 62 | Jev 1.13‡ Closed · EfficientLatency 0.243 sCost $0.029$28.85 per 1,000per 1M Jev 1.13 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://typesafe.ai/blog/introducing-system-one-models-and-jev, read 2026-10-08) Turkish grammar 46.4% (95%: 43.9–48.9%) Accuracy 57.5% (95%: 55.4–59.6%) Reading 83.7% · grammar 46.4% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 14.2%, 1 of them wrong Confidence: calibration gap 5.1 points (option score) Cost $0.029$28.85 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.243 s median per request, 1 in flight | 46.4%43.9%–48.9% | 57.5% | 83.7% | Not run | 5.1option score | $0.029$28.85at list prices | 0.2431 per request1 in flighton its repeats |
| 63 | d1‡ Closed · EfficientLatency 0.305 sCost $0.007$7.43 per 1,000per 1M d1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-18) Turkish grammar 46.0% (95%: 43.6–48.5%) Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.8% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 12.5%, 1 of them wrong Confidence: calibration gap 10.2 points (option score) Cost $0.007$7.43 per 1,000 questionsper 1M questions (free tier, at the list price) · 1 question per request Latency: 0.305 s median per request, 1 in flight | 46.0%43.6%–48.5% | 57.0% | 82.8% | Not run | 10.2option score | $0.007$7.43free tier, at the list price | 0.3051 per request1 in flighton its repeats |
| 64 | Jeffan-9B-1211‡ Open · 9.0BLatency 0.064 sCost $0.044$44.29 per 1,000per 1M Jeffan-9B-1211 · tattrongvu · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish grammar 46.0% (95%: 43.5–48.5%) Accuracy 57.0% (95%: 54.9–59.1%) Reading 82.9% · grammar 46.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.1% · error rate 42.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.6 points (option score) Cost $0.044$44.29 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.064 s median per request, 1 in flight | 46.0%43.5%–48.5% | 57.0% | 82.9% | 4.3%answered 98% | 4.6option score | $0.044$44.29from GPU time | 0.0641 per request1 in flight |
| 65 | Clef‡ Open · 27BLatency 0.205 sCost $0.110$110 per 1,000per 1M Clef · Cloudflare · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.48B). Turkish grammar 45.5% (95%: 43.0–48.0%) Accuracy 57.3% (95%: 55.2–59.3%) Reading 85.1% · grammar 45.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 7.4% · answered 96.8% · error rate 42.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.7 points (option score) Cost $0.110$110 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.205 s median per request, 1 in flight | 45.5%43.0%–48.0% | 57.3% | 85.1% | 7.4%answered 97% | 10.7option score | $0.110$110at the maker's list price | 0.2051 per request1 in flight |
| 66 | Drex 2.0‡ Open · 35B MoE · 3.0B activeLatency 0.325 sCost $0.019$19.15 per 1,000per 1M Drex 2.0 · Nace.AI · decision model Mixture of experts: 35B total (counted from the released files: 35.11B), about 3B active per step. Size class uses the total (35B). Turkish grammar 44.8% (95%: 42.3–47.3%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 81.4% · grammar 44.8% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence with no observed mistakes: 1.5% Confidence: calibration gap 4.2 points (option score) Cost $0.019$19.15 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.325 s median per request, 1 in flight | 44.8%42.3%–47.3% | 55.7% | 81.4% | Not run | 4.2option score | $0.019$19.15at list prices | 0.3251 per request1 in flight |
| 67 | Claude Haiku 5.5 max · chat‡ Closed · EfficientLatency 417.0 sCost $2.965$2,965 per 1,000per 1M Claude Haiku 5.5 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/anthropic/claude-haiku-5.5, read 2026-11-10) Turkish grammar 44.4% (95%: 41.9–46.8%) Accuracy 57.1% (95%: 55.0–59.1%) Reading 86.9% · grammar 44.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.0% · answered 67.7% · error rate 9.3% Confidence: calibration gap 17.5 points (written estimate) Cost $2.965$2,965 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 417.0 s median per request, 4 in flight | 44.4%41.9%–46.8% | 57.1% | 86.9% | 22.0%answered 68% | 17.5written estimate | $2.965$2,965at list prices | 417.016 per request4 in flight |
| 68 | Cygnet‡ Open · 12BLatency 0.036 sCost $0.039$39.39 per 1,000per 1M Cygnet · Blockbrain Labs · decision model Base model: Gemma 4 12B IT Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Turkish grammar 44.2% (95%: 41.7–46.7%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.8% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 3.2% · answered 98.5% · error rate 44.2% Kept at 99% confidence with no observed mistakes: 0.1% Confidence: calibration gap 13.8 points (option score) Cost $0.039$39.39 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.036 s median per request, 1 in flight | 44.2%41.7%–46.7% | 55.7% | 82.8% | 3.2%answered 99% | 13.8option score | $0.039$39.39at a host's list price for the same weights | 0.0361 per request1 in flight |
| 69 | Winnow-12B‡ Open · 12BLatency 0.059 sCost $0.041$41.40 per 1,000per 1M Winnow-12B · EldanRing · decision model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.91B). Turkish grammar 44.2% (95%: 41.7–46.7%) Accuracy 55.5% (95%: 53.4–57.5%) Reading 82.0% · grammar 44.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 4.3% · answered 98.0% · error rate 44.8% Kept at 99% confidence: 7.4%, 1 of them wrong Confidence: calibration gap 19.7 points (option score) Cost $0.041$41.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 44.2%41.7%–46.7% | 55.5% | 82.0% | 4.3%answered 98% | 19.7option score | $0.041$41.40from GPU time | 0.0591 per request1 in flight |
| 70 | Open-Jev-27B-v1.1‡ Open · 27BLatency 0.386 sCost $0.271$271 per 1,000per 1M Open-Jev-27B-v1.1 · Zefan Cai · decision model Base model: Qwen3.8-27B + adapter Parameters: 27B dense, as the maker states it (counted from the released files: 27.80B). Turkish grammar 44.1% (95%: 41.6–46.6%) Accuracy 55.7% (95%: 53.6–57.8%) Reading 82.9% · grammar 44.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.4% · answered 99.4% · error rate 44.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 24.5 points (option score) Cost $0.271$271 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.386 s median per request, 1 in flight | 44.1%41.6%–46.6% | 55.7% | 82.9% | 1.4%answered 99% | 24.5option score | $0.271$271from GPU time | 0.3861 per request1 in flighton its repeats |
| 71 | KnowLine-9B-Gen1‡ Open · 9.0BLatency 0.057 sCost $0.041$40.69 per 1,000per 1M KnowLine-9B-Gen1 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish grammar 43.8% (95%: 41.3–46.3%) Accuracy 54.5% (95%: 52.5–56.6%) Reading 79.9% · grammar 43.8% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.5% · answered 99.8% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 2.8% Confidence: calibration gap 4.6 points (option score) Cost $0.041$40.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.057 s median per request, 1 in flight | 43.8%41.3%–46.3% | 54.5% | 79.9% | 0.5%answered 100% | 4.6option score | $0.041$40.69from GPU time | 0.0571 per request1 in flight |
| 72 | Cerno 9B v1.2‡ Open · 9.0BLatency 0.051 sCost $0.036$36.02 per 1,000per 1M Cerno 9B v1.2 · Apertix AI · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish grammar 43.7% (95%: 41.3–46.2%) Accuracy 54.5% (95%: 52.4–56.5%) Reading 79.7% · grammar 43.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.2% · answered 99.9% · error rate 45.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.8 points (option score) Cost $0.036$36.02 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.051 s median per request, 1 in flight | 43.7%41.3%–46.2% | 54.5% | 79.7% | 0.2%answered 100% | 9.8option score | $0.036$36.02from GPU time | 0.0511 per request1 in flight |
| 73 | KnowLine-9B-Gen2‡ Open · 9.0BLatency 0.048 sCost $0.035$35.40 per 1,000per 1M KnowLine-9B-Gen2 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish grammar 42.1% (95%: 39.6–44.6%) Accuracy 53.3% (95%: 51.2–55.4%) Reading 79.6% · grammar 42.1% ‡ Turkish not officially supported: it lists other languages (source, read 16 November 2026) Non-hallucination with “I don't know” 2.6% · answered 98.8% · error rate 47.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.1 points (option score) Cost $0.035$35.40 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.048 s median per request, 1 in flight | 42.1%39.6%–44.6% | 53.3% | 79.6% | 2.6%answered 99% | 5.1option score | $0.035$35.40from GPU time | 0.0481 per request1 in flight |
| 74 | Adjudex 9B v5‡ Open · 9.0BLatency 0.038 sCost $0.026$26.30 per 1,000per 1M Adjudex 9B v5 · newsbang · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish grammar 41.4% (95%: 38.9–43.9%) Accuracy 53.1% (95%: 51.0–55.2%) Reading 80.8% · grammar 41.4% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.6% · error rate 46.6% Kept at 99% confidence: 6.0%, 2 of them wrong Confidence: calibration gap 9.2 points (option score) Cost $0.026$26.30 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.038 s median per request, 1 in flight | 41.4%38.9%–43.9% | 53.1% | 80.8% | 0.8%answered 100% | 9.2option score | $0.026$26.30from GPU time | 0.0381 per request1 in flight |
| 75 | Elephant-1‡ Open · 9.0BLatency 0.074 sCost $0.014$13.87 per 1,000per 1M Elephant-1 · Base Labs · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.41B). Turkish grammar 40.4% (95%: 38.0–42.9%) Accuracy 52.4% (95%: 50.3–54.4%) Reading 80.5% · grammar 40.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 34.8% · answered 81.6% · error rate 42.3% Kept at 99% confidence with no observed mistakes: 7.1% Confidence: calibration gap 11.8 points (option score) Cost $0.014$13.87 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.074 s median per request, 1 in flight | 40.4%38.0%–42.9% | 52.4% | 80.5% | 34.8%answered 82% | 11.8option score | $0.014$13.87at the maker's list price | 0.0741 per request1 in flight |
| 76 | GLiDE 9B‡ Open · 7.9BLatency 0.207 sCost No list price GLiDE 9B · Fastino · decision model Parameters: 7.94B dense, as the maker states it (counted from the released files: 7.94B). Turkish grammar 40.1% (95%: 37.7–42.5%) Accuracy 52.0% (95%: 50.0–54.1%) Reading 80.2% · grammar 40.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 24.1% · answered 87.4% · error rate 45.4% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 8.5 points (option score) Cost: No list price · 1 question per request Latency: 0.207 s median per request, 1 in flight | 40.1%37.7%–42.5% | 52.0% | 80.2% | 24.1%answered 87% | 8.5option score | No list price | 0.2071 per request1 in flight |
| 77 | Dexy-Decision-9B-V2‡ Open · 9.0BLatency 0.031 sCost $0.023$23.03 per 1,000per 1M Dexy-Decision-9B-V2 · Patrick-18 · decision model Base model: vllm-sr/d3-flash Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish grammar 39.7% (95%: 37.3–42.2%) Accuracy 51.6% (95%: 49.5–53.7%) Reading 79.6% · grammar 39.7% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 3.4% · answered 98.3% · error rate 48.4% Kept at 99% confidence: 6.3%, 1 of them wrong Confidence: calibration gap 9.1 points (option score) Cost $0.023$23.03 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.031 s median per request, 1 in flight | 39.7%37.3%–42.2% | 51.6% | 79.6% | 3.4%answered 98% | 9.1option score | $0.023$23.03from GPU time | 0.0311 per request1 in flight |
| 78 | StartLux-Decision-v1.1-9B-Instant‡ Open · 9.0BLatency 0.154 sCost No list price StartLux-Decision-v1.1-9B-Instant · StartLux · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish grammar 39.5% (95%: 37.1–42.0%) Accuracy 51.2% (95%: 49.1–53.3%) Reading 78.8% · grammar 39.5% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 2.2% · answered 98.9% · error rate 49.2% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 7.6 points (option score) Cost: No list price · 1 question per request Latency: 0.154 s median per request, 1 in flight | 39.5%37.1%–42.0% | 51.2% | 78.8% | 2.2%answered 99% | 7.6option score | No list price | 0.1541 per request1 in flight |
| 79 | Perplexity Decider Flash v1.3‡ Open · 4.0BLatency 0.033 sCost $0.025$24.69 per 1,000per 1M Perplexity Decider Flash v1.3 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish grammar 39.2% (95%: 36.8–41.7%) Accuracy 49.7% (95%: 47.6–51.8%) Reading 74.4% · grammar 39.2% ‡ Turkish not officially supported: it names no language (source, read 14 November 2026) Non-hallucination with “I don't know” 6.5% · answered 96.6% · error rate 50.4% Kept at 99% confidence with no observed mistakes: 1.3% Confidence: calibration gap 12.6 points (option score) Cost $0.025$24.69 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.033 s median per request, 1 in flight | 39.2%36.8%–41.7% | 49.7% | 74.4% | 6.5%answered 97% | 12.6option score | $0.025$24.69from GPU time | 0.0331 per request1 in flight |
| 80 | vera-medium-omni‡ Open · 4.2BLatency 0.029 sCost $0.021$20.72 per 1,000per 1M vera-medium-omni · scar-ai · decision model Base model: Qwen3.5-4B Parameters: 4.2B dense, as the maker states it (counted from the released files: 4.21B). Turkish grammar 37.9% (95%: 35.5–40.3%) Accuracy 49.1% (95%: 47.0–51.2%) Reading 75.5% · grammar 37.9% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 8.9% · answered 95.2% · error rate 51.4% Kept at 99% confidence with no observed mistakes: 0.2% Confidence: calibration gap 3.7 points (option score) Cost $0.021$20.72 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.029 s median per request, 1 in flight | 37.9%35.5%–40.3% | 49.1% | 75.5% | 8.9%answered 95% | 3.7option score | $0.021$20.72from GPU time | 0.0291 per request1 in flight |
| 81 | KnowLine 4B Gen5.5‡ Open · 4.0BLatency 0.099 sCost No list price KnowLine 4B Gen5.5 · PelaAI · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish grammar 37.3% (95%: 34.9–39.7%) Accuracy 48.5% (95%: 46.4–50.6%) Reading 74.8% · grammar 37.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 4.7% · answered 97.5% · error rate 52.3% Kept at 99% confidence with no observed mistakes: 1.8% Confidence: calibration gap 13.5 points (option score) Cost: No list price · 1 question per request Latency: 0.099 s median per request, 1 in flight | 37.3%34.9%–39.7% | 48.5% | 74.8% | 4.7%answered 97% | 13.5option score | No list price | 0.0991 per request1 in flight |
| 82 | Perplexity Decider Flash v1.2‡ Open · 4.0BLatency 0.132 sCost $0.004$3.88 per 1,000per 1M Perplexity Decider Flash v1.2 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish grammar 37.2% (95%: 34.8–39.6%) Accuracy 47.8% (95%: 45.7–49.9%) Reading 72.7% · grammar 37.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.7% · answered 91.8% · error rate 51.7% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost $0.004$3.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.132 s median per request, 1 in flight | 37.2%34.8%–39.6% | 47.8% | 72.7% | 14.7%answered 92% | 8.3option score | $0.004$3.88at the maker's list price | 0.1321 per request1 in flight |
| 83 | Qwen3.5-4B reasoning off · chat‡ Open · 4.0BLatency 32.2 sCost $0.894$894 per 1,000per 1M Qwen3.5-4B reasoning off · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish grammar 36.7% (95%: 34.3–39.1%) Accuracy 46.2% (95%: 44.1–48.3%) Reading 68.4% · grammar 36.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 6.9% · answered 95.4% · error rate 54.0% Confidence: not measured Cost $0.894$894 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 32.2 s median per request, 32 in flight | 36.7%34.3%–39.1% | 46.2% | 68.4% | 6.9%answered 95% | Not measureda plain answer line, no confidence | $0.894$894from GPU time | 32.21 per request32 in flight |
| 84 | StartLux-Decision-v1.1-4B-Instant‡ Open · 4.0BLatency 0.074 sCost No list price StartLux-Decision-v1.1-4B-Instant · StartLux · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish grammar 36.7% (95%: 34.3–39.1%) Accuracy 47.9% (95%: 45.8–50.0%) Reading 74.1% · grammar 36.7% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 1.8% · answered 99.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.4% Confidence: calibration gap 7.1 points (option score) Cost: No list price · 1 question per request Latency: 0.074 s median per request, 1 in flight | 36.7%34.3%–39.1% | 47.9% | 74.1% | 1.8%answered 99% | 7.1option score | No list price | 0.0741 per request1 in flight |
| 85 | Chinese Decision Pointer 4B‡ Open · 4.0BLatency 0.035 sCost $0.025$24.58 per 1,000per 1M Chinese Decision Pointer 4B · quincyqiang · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish grammar 36.4% (95%: 34.0–38.8%) Accuracy 46.7% (95%: 44.6–48.8%) Reading 70.9% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 11.2% · answered 93.8% · error rate 52.5% Kept at 99% confidence: 2.3%, 1 of them wrong Confidence: calibration gap 11.7 points (option score) Cost $0.025$24.58 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.035 s median per request, 1 in flight | 36.4%34.0%–38.8% | 46.7% | 70.9% | 11.2%answered 94% | 11.7option score | $0.025$24.58from GPU time | 0.0351 per request1 in flight |
| 86 | Verdin-4B-v1.1‡ Open · 4.0BLatency 0.049 sCost $0.034$34.21 per 1,000per 1M Verdin-4B-v1.1 · victorbennett · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish grammar 36.4% (95%: 34.0–38.8%) Accuracy 46.8% (95%: 44.7–48.9%) Reading 71.2% · grammar 36.4% ‡ Turkish not officially supported: it lists other languages (source, read 15 November 2026) Non-hallucination with “I don't know” 0.4% · answered 99.8% · error rate 53.3% Kept at 99% confidence with no observed mistakes: 1.2% Confidence: calibration gap 9.8 points (option score) Cost $0.034$34.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.049 s median per request, 1 in flight | 36.4%34.0%–38.8% | 46.8% | 71.2% | 0.4%answered 100% | 9.8option score | $0.034$34.21from GPU time | 0.0491 per request1 in flight |
| 87 | Valen-Decision-1.5-9B‡ Open · 9.0BLatency 0.059 sCost $0.040$40.49 per 1,000per 1M Valen-Decision-1.5-9B · Valen-Team · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.40B). Turkish grammar 35.0% (95%: 32.7–37.4%) Accuracy 47.0% (95%: 44.9–49.0%) Reading 75.0% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 0.1% · answered 100.0% · error rate 52.7% Kept at 99% confidence with no observed mistakes: 0.3% Confidence: calibration gap 3.8 points (option score) Cost $0.040$40.49 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.059 s median per request, 1 in flight | 35.0%32.7%–37.4% | 47.0% | 75.0% | 0.1%answered 100% | 3.8option score | $0.040$40.49from GPU time | 0.0591 per request1 in flight |
| 88 | Microsoft-Decision-1‡ Closed · EfficientLatency 0.286 sCost $0.014$13.82 per 1,000per 1M Microsoft-Decision-1 · Microsoft · decision model Base model: Qwen3.5-9B Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://openrouter.ai/microsoft/microsoft-decision-1, read 2026-10-17) Turkish grammar 35.0% (95%: 32.6–37.4%) Accuracy 46.5% (95%: 44.4–48.5%) Reading 73.5% · grammar 35.0% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: 2.8%, 3 of them wrong Confidence: calibration gap 19.8 points (option score) Cost $0.014$13.82 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.286 s median per request, 1 in flight | 35.0%32.6%–37.4% | 46.5% | 73.5% | Not run | 19.8option score | $0.014$13.82at list prices | 0.2861 per request1 in flight |
| 89 | Qwen3-14B reasoning off · chat‡ Open · 15BLatency 2.85 sCost $0.032$32.31 per 1,000per 1M Qwen3-14B reasoning off · chat · Alibaba · chat model Parameters: 14.8B dense, as the maker states it (counted from the released files: 14.77B). Turkish grammar 34.5% (95%: 32.2–36.9%) Accuracy 45.7% (95%: 43.6–47.8%) Reading 72.1% · grammar 34.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 1.9% · answered 99.0% · error rate 53.8% Confidence: calibration gap 41.0 points (answer letter probability) Cost $0.032$32.31 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 2.85 s median per request, 1 in flight | 34.5%32.2%–36.9% | 45.7% | 72.1% | 1.9%answered 99% | 41.0answer letter probability | $0.032$32.31at a host's list price for the same weights | 2.8516 per request1 in flight |
| 90 | Clef-Flash‡ Open · 9.0BLatency 0.075 sCost $0.017$17.37 per 1,000per 1M Clef-Flash · Cloudflare · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.53B). Turkish grammar 34.1% (95%: 31.8–36.5%) Accuracy 45.7% (95%: 43.6–47.8%) Reading 73.0% · grammar 34.1% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.0% · answered 95.0% · error rate 53.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 15.5 points (option score) Cost $0.017$17.37 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.075 s median per request, 1 in flight | 34.1%31.8%–36.5% | 45.7% | 73.0% | 9.0%answered 95% | 15.5option score | $0.017$17.37at the maker's list price | 0.0751 per request1 in flight |
| 91 | Erk-14B reasoning off · chat Open · 14BLatency 2.39 sCost $0.106$106 per 1,000per 1M Erk-14B reasoning off · chat · eCloud Technology · chat model Parameters: 14B dense, as the maker states it (counted from the released files: 14.77B). Turkish grammar 33.4% (95%: 31.1–35.8%) Accuracy 45.0% (95%: 42.9–47.1%) Reading 72.3% · grammar 33.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.3% · answered 99.8% · error rate 55.3% Confidence: calibration gap 13.5 points (answer letter probability) Cost $0.106$106 per 1,000 questionsper 1M questions (from GPU time) · 16 questions per request Latency: 2.39 s median per request, 1 in flight | 33.4%31.1%–35.8% | 45.0% | 72.3% | 0.3%answered 100% | 13.5answer letter probability | $0.106$106from GPU time | 2.3916 per request1 in flight |
| 92 | Metask-Jev-4B Open · 4.5BLatency 0.090 sCost $0.063$62.82 per 1,000per 1M Metask-Jev-4B · Metask Lab · decision model Parameters: 4.54B dense, as the maker states it (counted from the released files: 4.54B). Turkish grammar 31.4% (95%: 29.1–33.7%) Accuracy 40.8% (95%: 38.7–42.8%) Reading 62.8% · grammar 31.4% Turkish: listed in its card (source, read 12 November 2026) Non-hallucination with “I don't know” 0.8% · answered 99.5% · error rate 60.1% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.063$62.82 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.090 s median per request, 1 in flight | 31.4%29.1%–33.7% | 40.8% | 62.8% | 0.8%answered 100% | 5.8option score | $0.063$62.82from GPU time | 0.0901 per request1 in flight |
| 93 | KnowLine-2B-Gen1‡ Open · 2.0BLatency 0.062 sCost No list price KnowLine-2B-Gen1 · PelaAI · decision model Base model: Qwen3.5-2B Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish grammar 29.4% (95%: 27.2–31.7%) Accuracy 36.8% (95%: 34.8–38.8%) Reading 54.3% · grammar 29.4% ‡ Turkish not officially supported: it lists other languages (source, read 13 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.4 points (option score) Cost: No list price · 1 question per request Latency: 0.062 s median per request, 1 in flight | 29.4%27.2%–31.7% | 36.8% | 54.3% | Not run | 8.4option score | No list price | 0.0621 per request1 in flight |
| 94 | LFM2.5-2.6B default reasoning · chat‡ Open · 2.7BLatency 51.2 sCost $2.761$2,761 per 1,000per 1M LFM2.5-2.6B default reasoning · chat · Liquid AI · chat model Parameters: 2.69B dense, as the maker states it (counted from the released files: 2.70B). Turkish grammar 29.0% (95%: 26.8–31.3%) Accuracy 33.9% (95%: 31.9–35.9%) Reading 45.4% · grammar 29.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 24.7% · answered 81.6% · error rate 63.2% Confidence: not measured Cost $2.761$2,761 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 51.2 s median per request, 32 in flight | 29.0%26.8%–31.3% | 33.9% | 45.4% | 24.7%answered 82% | Not measureda plain answer line, no confidence | $2.761$2,761at the maker's list price | 51.21 per request32 in flight |
| 95 | GLiDE 2B‡ Open · 1.9BLatency 0.034 sCost No list price GLiDE 2B · Fastino · decision model Parameters: 1.88B dense, as the maker states it (counted from the released files: 1.88B). Turkish grammar 28.5% (95%: 26.3–30.8%) Accuracy 35.3% (95%: 33.3–37.3%) Reading 51.2% · grammar 28.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 38.5% · answered 72.2% · error rate 61.7% Kept at 99% confidence with no observed mistakes: 0.9% Confidence: calibration gap 20.7 points (option score) Cost: No list price · 1 question per request Latency: 0.034 s median per request, 1 in flight | 28.5%26.3%–30.8% | 35.3% | 51.2% | 38.5%answered 72% | 20.7option score | No list price | 0.0341 per request1 in flight |
| 96 | LFM3-1.7B reasoning on · chat‡ Open · 1.7BLatency 24.2 sCost $1.277$1,277 per 1,000per 1M LFM3-1.7B reasoning on · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish grammar 27.4% (95%: 25.2–29.6%) Accuracy 32.4% (95%: 30.5–34.4%) Reading 44.2% · grammar 27.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 67.0% · answered 45.5% · error rate 58.8% Confidence: not measured Cost $1.277$1,277 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 24.2 s median per request, 32 in flight | 27.4%25.2%–29.6% | 32.4% | 44.2% | 67.0%answered 46% | Not measureda plain answer line, no confidence | $1.277$1,277at the maker's list price | 24.21 per request32 in flight |
| 97 | StartLux-Decision-v1.1-2B-Instant‡ Open · 2.0BLatency 0.040 sCost No list price StartLux-Decision-v1.1-2B-Instant · StartLux · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish grammar 26.3% (95%: 24.2–28.6%) Accuracy 34.3% (95%: 32.3–36.3%) Reading 52.9% · grammar 26.3% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 21.5% · answered 84.8% · error rate 65.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 6.4 points (option score) Cost: No list price · 1 question per request Latency: 0.040 s median per request, 1 in flight | 26.3%24.2%–28.6% | 34.3% | 52.9% | 21.5%answered 85% | 6.4option score | No list price | 0.0401 per request1 in flight |
| 98 | StartLux-Decision-v1.1-0.8B-Instant‡ Open · 0.8BLatency 0.028 sCost No list price StartLux-Decision-v1.1-0.8B-Instant · StartLux · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish grammar 25.8% (95%: 23.7–28.1%) Accuracy 28.6% (95%: 26.7–30.5%) Reading 35.1% · grammar 25.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 12.4% · answered 90.9% · error rate 70.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 4.4 points (option score) Cost: No list price · 1 question per request Latency: 0.028 s median per request, 1 in flight | 25.8%23.7%–28.1% | 28.6% | 35.1% | 12.4%answered 91% | 4.4option score | No list price | 0.0281 per request1 in flight |
| 99 | OneJev1.6-0.8B‡ Open · 0.8BLatency 0.017 sCost No list price OneJev1.6-0.8B · OmniJev · decision model Parameters: 800M dense, as the maker states it (counted from the released files: 0.85B). Turkish grammar 24.6% (95%: 22.6–26.9%) Accuracy 29.1% (95%: 27.3–31.1%) Reading 39.6% · grammar 24.6% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 43.4% · answered 65.7% · error rate 68.2% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 9.3 points (option score) Cost: No list price · 1 question per request Latency: 0.017 s median per request, 1 in flight | 24.6%22.6%–26.9% | 29.1% | 39.6% | 43.4%answered 66% | 9.3option score | No list price | 0.0171 per request1 in flight |
| 100 | LFM3-1.7B reasoning off · chat‡ Open · 1.7BLatency 4.90 sCost $0.646$646 per 1,000per 1M LFM3-1.7B reasoning off · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish grammar 24.6% (95%: 22.5–26.8%) Accuracy 28.3% (95%: 26.5–30.3%) Reading 37.2% · grammar 24.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 76.2% · answered 32.7% · error rate 62.4% Confidence: not measured Cost $0.646$646 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 4.90 s median per request, 32 in flight | 24.6%22.5%–26.8% | 28.3% | 37.2% | 76.2%answered 33% | Not measureda plain answer line, no confidence | $0.646$646at the maker's list price | 4.901 per request32 in flight |
| 101 | d1-2B‡ Open · 1.9BLatency 0.019 sCost $0.013$13.17 per 1,000per 1M d1-2B · Liquid AI · decision model Base model: LFM3-VL-2B Parameters: 1.92B dense, as the maker states it (counted from the released files: 1.92B). Turkish grammar 24.5% (95%: 22.4–26.7%) Accuracy 29.4% (95%: 27.5–31.3%) Reading 40.9% · grammar 24.5% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 65.5% · answered 45.0% · error rate 64.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 5.8 points (option score) Cost $0.013$13.17 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.019 s median per request, 1 in flight | 24.5%22.4%–26.7% | 29.4% | 40.9% | 65.5%answered 45% | 5.8option score | $0.013$13.17from GPU time | 0.0191 per request1 in flight |
| 102 | Decider 1‡ Closed · EfficientLatency 0.382 sCost $0.015$14.76 per 1,000per 1M Decider 1 · meraGPT · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://meragpt.com/docs/models, read 2026-10-19) Turkish grammar 23.9% (95%: 21.9–26.1%) Accuracy 27.1% (95%: 25.3–29.0%) Reading 34.6% · grammar 23.9% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 14.3 points (option score) Cost $0.015$14.76 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.382 s median per request, 1 in flight | 23.9%21.9%–26.1% | 27.1% | 34.6% | Not run | 14.3option score | $0.015$14.76at list prices | 0.3821 per request1 in flight |
| 103 | GLiDE 0.8B‡ Open · 0.8BLatency 0.022 sCost No list price GLiDE 0.8B · Fastino · decision model Parameters: 750M dense, as the maker states it (counted from the released files: 0.75B). Turkish grammar 23.7% (95%: 21.6–25.9%) Accuracy 29.3% (95%: 27.4–31.2%) Reading 42.4% · grammar 23.7% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 29.2% · answered 78.2% · error rate 67.7% Kept at 99% confidence: 0.8%, 1 of them wrong Confidence: calibration gap 24.8 points (option score) Cost: No list price · 1 question per request Latency: 0.022 s median per request, 1 in flight | 23.7%21.6%–25.9% | 29.3% | 42.4% | 29.2%answered 78% | 24.8option score | No list price | 0.0221 per request1 in flight |
| 104 | Strands Decider 2B‡ Open · 2.0BLatency 0.026 sCost $0.020$20.12 per 1,000per 1M Strands Decider 2B · Strands Agents · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.29B). Turkish grammar 23.5% (95%: 21.4–25.7%) Accuracy 27.5% (95%: 25.7–29.4%) Reading 36.9% · grammar 23.5% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 9.2% · answered 93.2% · error rate 72.4% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 12.2 points (option score) Cost $0.020$20.12 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.026 s median per request, 1 in flight | 23.5%21.4%–25.7% | 27.5% | 36.9% | 9.2%answered 93% | 12.2option score | $0.020$20.12from GPU time | 0.0261 per request1 in flight |
| 105 | KnowLine-0.8B-Gen1‡ Open · 0.8BLatency 0.033 sCost No list price KnowLine-0.8B-Gen1 · PelaAI · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish grammar 22.4% (95%: 20.4–24.5%) Accuracy 27.1% (95%: 25.3–29.0%) Reading 38.3% · grammar 22.4% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 95.3% · error rate 72.3% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.3 points (option score) Cost: No list price · 1 question per request Latency: 0.033 s median per request, 1 in flight | 22.4%20.4%–24.5% | 27.1% | 38.3% | 6.4%answered 95% | 8.3option score | No list price | 0.0331 per request1 in flight |
| 106 | GLiClass v4 Edge‡ Open · 0.4BLatency 0.025 sCost No list price GLiClass v4 Edge · Knowledgator · decision model Base model: DeBERTa-v3-large Parameters: 400M dense, as the maker states it (counted from the released files: 0.44B). Turkish grammar 22.2% (95%: 20.2–24.4%) Accuracy 23.3% (95%: 21.6–25.2%) Reading 25.9% · grammar 22.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 22.3% · answered 81.9% · error rate 77.0% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 10.5 points (option score) Cost: No list price · 1 question per request Latency: 0.025 s median per request, 1 in flight | 22.2%20.2%–24.4% | 23.3% | 25.9% | 22.3%answered 82% | 10.5option score | No list price | 0.0251 per request1 in flight |
| 107 | Qwen3.5-2B reasoning off · chat‡ Open · 2.0BLatency 8.84 sCost $0.268$268 per 1,000per 1M Qwen3.5-2B reasoning off · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish grammar 21.6% (95%: 19.6–23.7%) Accuracy 25.4% (95%: 23.6–27.2%) Reading 34.3% · grammar 21.6% ‡ Turkish not officially supported: it names no language (source, read 15 November 2026) Non-hallucination with “I don't know” 20.8% · answered 75.6% · error rate 72.3% Confidence: not measured Cost $0.268$268 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 8.84 s median per request, 32 in flight | 21.6%19.6%–23.7% | 25.4% | 34.3% | 20.8%answered 76% | Not measureda plain answer line, no confidence | $0.268$268from GPU time | 8.841 per request32 in flight |
| 108 | GLiDE 0.4B‡ Open · 0.4BLatency 0.016 sCost No list price GLiDE 0.4B · Fastino · decision model Parameters: 400M dense, as the maker states it (counted from the released files: 0.40B). Turkish grammar 21.3% (95%: 19.3–23.4%) Accuracy 22.2% (95%: 20.6–24.0%) Reading 24.5% · grammar 21.3% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 95.3% · answered 5.8% · error rate 78.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 8.0 points (option score) Cost: No list price · 1 question per request Latency: 0.016 s median per request, 1 in flight | 21.3%19.3%–23.4% | 22.2% | 24.5% | 95.3%answered 6% | 8.0option score | No list price | 0.0161 per request1 in flight |
| 109 | GLiNER2.5-multi-Decide‡ Open · 0.3BLatency 0.009 sCost $0.006$6.27 per 1,000per 1M GLiNER2.5-multi-Decide · Fastino · decision model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish grammar 19.8% (95%: 17.9–21.9%) Accuracy 20.2% (95%: 18.6–21.9%) Reading 21.0% · grammar 19.8% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.4% · answered 94.8% · error rate 79.9% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 23.1 points (option score) Cost $0.006$6.27 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 19.8%17.9%–21.9% | 20.2% | 21.0% | 6.4%answered 95% | 23.1option score | $0.006$6.27from GPU time | 0.0091 per request1 in flight |
| 110 | LFM2.5-8B-A1B default reasoning · chat‡ Open · 8.3B MoE · 1.5B activeLatency 39.2 sCost $1.050$1,050 per 1,000per 1M LFM2.5-8B-A1B default reasoning · chat · Liquid AI · chat model Mixture of experts: 8.3B total (counted from the released files: 8.47B), about 1.5B active per step. Size class uses the total (8.3B). Turkish grammar 19.0% (95%: 17.1–21.0%) Accuracy 21.0% (95%: 19.3–22.7%) Reading 25.6% · grammar 19.0% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 56.0% · answered 42.4% · error rate 74.0% Confidence: not measured Cost $1.050$1,050 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 39.2 s median per request, 32 in flight | 19.0%17.1%–21.0% | 21.0% | 25.6% | 56.0%answered 42% | Not measureda plain answer line, no confidence | $1.050$1,050from GPU time | 39.21 per request32 in flight |
| 111 | Laya Multilingual‡ Open · 0.3BLatency 0.009 sCost $0.008$7.74 per 1,000per 1M Laya Multilingual · Convai Innovations · decision model Parameters: 322M dense, as the maker states it (counted from the released files: 0.32B). Turkish grammar 18.2% (95%: 16.4–20.2%) Accuracy 17.9% (95%: 16.3–19.5%) Reading 17.1% · grammar 18.2% ‡ Turkish not officially supported: it names no language (source, read 12 November 2026) Non-hallucination with “I don't know” 14.8% · answered 87.3% · error rate 83.8% Kept at 99% confidence: 0.0%, 1 of them wrong Confidence: calibration gap 27.7 points (option score) Cost $0.008$7.74 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.009 s median per request, 1 in flight | 18.2%16.4%–20.2% | 17.9% | 17.1% | 14.8%answered 87% | 27.7option score | $0.008$7.74from GPU time | 0.0091 per request1 in flighton its repeats |
| 112 | Laya‡ Open · 0.4BLatency 0.011 sCost $0.009$9.00 per 1,000per 1M Laya · Convai Innovations · decision model Parameters: 421M dense, as the maker states it (counted from the released files: 0.42B). Turkish grammar 18.2% (95%: 16.4–20.2%) Accuracy 19.0% (95%: 17.4–20.7%) Reading 20.9% · grammar 18.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 28.3% · answered 75.6% · error rate 81.5% Kept at 99% confidence: No answers at this threshold Confidence: calibration gap 13.2 points (option score) Cost $0.009$9.00 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 0.011 s median per request, 1 in flight | 18.2%16.4%–20.2% | 19.0% | 20.9% | 28.3%answered 76% | 13.2option score | $0.009$9.00from GPU time | 0.0111 per request1 in flighton its repeats |
| 113 | GLiNER2.5 Multi‡ Open · 0.3BLatency 0.032 sCost $0.006$5.88 per 1,000per 1M GLiNER2.5 Multi · Fastino · chat model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish grammar 17.6% (95%: 15.8–19.6%) Accuracy 17.5% (95%: 15.9–19.1%) Reading 17.2% · grammar 17.6% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 6.3% · answered 94.7% · error rate 82.6% Confidence: calibration gap 35.6 points (option score) Cost $0.006$5.88 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.032 s median per request, 1 in flight | 17.6%15.8%–19.6% | 17.5% | 17.2% | 6.3%answered 95% | 35.6option score | $0.006$5.88at the maker's list price | 0.0321 per request1 in flight |
| 114 | MiniCPM5-2B reasoning off · chat‡ Open · 2.5BLatency 172.0 sCost $5.464$5,464 per 1,000per 1M MiniCPM5-2B reasoning off · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Turkish grammar 5.2% (95%: 4.2–6.4%) Accuracy 5.6% (95%: 4.7–6.6%) Reading 6.6% · grammar 5.2% ‡ Turkish not officially supported: it lists other languages (source, read 12 November 2026) Non-hallucination with “I don't know” 25.4% · answered 20.1% · error rate 75.3% Confidence: not measured Cost $5.464$5,464 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Latency: 172.0 s median per request, 32 in flight | 5.2%4.2%–6.4% | 5.6% | 6.6% | 25.4%answered 20% | Not measureda plain answer line, no confidence | $5.464$5,464from GPU time | 172.01 per request32 in flight |
Accuracy on grammar: the share of the 1,542 Turkish grammar questions answered right, with its 95% interval as the whisker; higher is better.
No row of this part matches these keys. The small open chat models ran in English only.
Rank: the row's place among the rows in view, by accuracy, 1, 2, 3 without repeats; rows with the same score keep the table's order, which claims no difference between them. No all-pairs test covers the Turkish grammar part, so it has no tiers. A setup with web search is not ranked (Rank: –).
‡ Turkish not officially supported: the model's own card or maker docs do not list Turkish (read 16 November 2026; Ümit #5067); of the models here only Erk-14B and Metask-Jev-4B list it.
| Rank | Model | Accuracy in Englishof 547 questions, 95% interval; min–max of 3 seeds where marked | Non-hallucination% “I don't know” of answers not right | Confidencechosen-answer Brier | Costper 1,000 questionsper 1M questions | Latencymedian s per request |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 max · chat Closed · FlagshipLatency 74.0 sCost $52.942$52,942 per 1,000per 1M Claude Fable 5.2 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $8.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-11-11) Accuracy in English 98.0% (95%: 96.4–98.9%), 547 kept reading questions Non-hallucination with “I don't know” 28.6% · answered 99.3% · error rate 1.8% Cost $52.942$52,942 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 74.0 s median per request, 1 in flight English runs on: hosted API | 98.0%96.4%–98.9% | 28.6%answered 99% | Not measured | $52.942$52,942subscription, priced as the API | 74.016 per request1 in flight |
| 2 | Gemini 4 Hydrogen high · chat Closed · WorkhorseLatency 28.4 sCost $4.927$4,927 per 1,000per 1M Gemini 4 Hydrogen high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Accuracy in English 96.7% (95%: 94.9–97.9%), 547 kept reading questions Non-hallucination with “I don't know” 20.0% · answered 99.3% · error rate 2.9% Cost $4.927$4,927 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 28.4 s median per request, 1 in flight English runs on: hosted API | 96.7%94.9%–97.9% | 20.0%answered 99% | Not measured | $4.927$4,927subscription, priced as the API | 28.416 per request1 in flight |
| 3 | GPT-6.1 Terra high · chat Closed · WorkhorseLatency 7.94 sCost $0.509$509 per 1,000per 1M GPT-6.1 Terra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Accuracy in English 96.2% (95%: 94.2–97.5%), 547 kept reading questions Non-hallucination with “I don't know” 16.7% · answered 99.3% · error rate 3.7% Cost $0.509$509 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.94 s median per request, 1 in flight English runs on: hosted API | 96.2%94.2%–97.5% | 16.7%answered 99% | Not measured | $0.509$509subscription, priced as the API | 7.9416 per request1 in flight |
| GPT-6.1 Terra low · chat Closed · Workhorseother setting of GPT-6.1 TerraLatency 7.29 sCost $0.491$491 per 1,000per 1M GPT-6.1 Terra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Accuracy in English 96.2% (95%: 94.2–97.5%), 547 kept reading questions Non-hallucination with “I don't know” 15.4% · answered 99.3% · error rate 4.1% Cost $0.491$491 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.29 s median per request, 1 in flight English runs on: hosted API | 96.2%94.2%–97.5% | 15.4%answered 99% | Not measured | $0.491$491subscription, priced as the API | 7.2916 per request1 in flight | |
| 4 | GPT-6 Astra high · chat Closed · FlagshipLatency 15.6 sCost $6.334$6,334 per 1,000per 1M GPT-6 Astra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 96.2% (95%: 94.2–97.5%), 547 kept reading questions Non-hallucination with “I don't know” 19.0% · answered 99.3% · error rate 3.1% Cost $6.334$6,334 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 15.6 s median per request, 1 in flight English runs on: hosted API | 96.2%94.2%–97.5% | 19.0%answered 99% | Not measured | $6.334$6,334subscription, priced as the API | 15.616 per request1 in flight |
| GPT-6 Astra medium · chat Closed · Flagshipother setting of GPT-6 AstraLatency 10.7 sCost $4.696$4,696 per 1,000per 1M GPT-6 Astra medium · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 96.0% (95%: 94.0–97.3%), 547 kept reading questions Non-hallucination with “I don't know” 16.7% · answered 99.3% · error rate 3.7% Cost $4.696$4,696 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 10.7 s median per request, 1 in flight English runs on: hosted API | 96.0%94.0%–97.3% | 16.7%answered 99% | Not measured | $4.696$4,696subscription, priced as the API | 10.716 per request1 in flight | |
| GPT-6 Astra low · chat Closed · Flagshipother setting of GPT-6 AstraLatency 11.5 sCost $5.918$5,918 per 1,000per 1M GPT-6 Astra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 96.3% (95%: 94.4–97.6%), 547 kept reading questions Non-hallucination with “I don't know” 18.2% · answered 99.3% · error rate 3.3% Cost $5.918$5,918 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 11.5 s median per request, 1 in flight English runs on: hosted API | 96.3%94.4%–97.6% | 18.2%answered 99% | Not measured | $5.918$5,918subscription, priced as the API | 11.516 per request1 in flight | |
| GPT-6 Astra low · chat · 1 per request Closed · Flagshipother setting of GPT-6 AstraLatency 6.43 sCost $28.653$28,653 per 1,000per 1M GPT-6 Astra low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 95.6% (95%: 93.6–97.0%), 547 kept reading questions Non-hallucination with “I don't know” 14.8% · answered 99.3% · error rate 4.2% Cost $28.653$28,653 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 6.43 s median per request, 1 in flight English runs on: hosted API | 95.6%93.6%–97.0% | 14.8%answered 99% | Not measured | $28.653$28,653subscription, priced as the API | 6.431 per request1 in flight | |
| 5 | Gemini 3.8 Flash high · chat Closed · WorkhorseLatency 104.6 sCost $3.695$3,695 per 1,000per 1M Gemini 3.8 Flash high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy in English 96.0% (95%: 94.0–97.3%), 547 kept reading questions Non-hallucination with “I don't know” 10.0% · answered 99.6% · error rate 3.3% Cost $3.695$3,695 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 104.6 s median per request, 4 in flight English runs on: hosted API | 96.0%94.0%–97.3% | 10.0%answered 100% | Not measured | $3.695$3,695at list prices | 104.616 per request4 in flight |
| Gemini 3.8 Flash low · chat Closed · Workhorseother setting of Gemini 3.8 FlashLatency 4.19 sCost $0.288$288 per 1,000per 1M Gemini 3.8 Flash low · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy in English 94.3% (95%: 92.1–96.0%), 547 kept reading questions Non-hallucination with “I don't know” 3.4% · answered 99.8% · error rate 5.1% Cost $0.288$288 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 4.19 s median per request, 1 in flight English runs on: hosted API | 94.3%92.1%–96.0% | 3.4%answered 100% | Not measured | $0.288$288at list prices | 4.1916 per request1 in flight | |
| Gemini 3.8 Flash low · chat · 1 per request Closed · Workhorseother setting of Gemini 3.8 FlashLatency 2.85 sCost $0.858$858 per 1,000per 1M Gemini 3.8 Flash low · chat · 1 per request · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Accuracy in English 95.2% (95%: 93.1–96.7%), 547 kept reading questions Non-hallucination with “I don't know” 7.7% · answered 99.6% · error rate 4.4% Cost $0.858$858 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 2.85 s median per request, 1 in flight English runs on: hosted API | 95.2%93.1%–96.7% | 7.7%answered 100% | Not measured | $0.858$858at list prices | 2.851 per request1 in flight | |
| 6 | Claude Fable 5.1 high · chat Closed · FlagshipLatency 15.6 sCost $11.531$11,531 per 1,000per 1M Claude Fable 5.1 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-28) Accuracy in English 96.0% (95%: 94.0–97.3%), 547 kept reading questions Non-hallucination with “I don't know” 7.4% · answered 99.6% · error rate 4.6% Cost $11.531$11,531 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 15.6 s median per request, 1 in flight English runs on: hosted API | 96.0%94.0%–97.3% | 7.4%answered 100% | Not measured | $11.531$11,531subscription, priced as the API | 15.616 per request1 in flight |
| 7 | Gemini 4 Argon low · chat Closed · FlagshipLatency 6.03 sCost $1.900$1,900 per 1,000per 1M Gemini 4 Argon low · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-28) Accuracy in English 96.0% (95%: 94.0–97.3%), 547 kept reading questions Non-hallucination with “I don't know” 18.5% · answered 99.1% · error rate 4.1% Cost $1.900$1,900 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 6.03 s median per request, 1 in flight English runs on: hosted API | 96.0%94.0%–97.3% | 18.5%answered 99% | Not measured | $1.900$1,900at list prices | 6.0316 per request1 in flight |
| 8 | GPT-6.1 Sol high · chat Closed · WorkhorseLatency 16.9 sCost $1.427$1,427 per 1,000per 1M GPT-6.1 Sol high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 95.6% (95%: 93.6–97.0%), 547 kept reading questions Non-hallucination with “I don't know” 16.7% · answered 99.3% · error rate 3.7% Cost $1.427$1,427 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 16.9 s median per request, 1 in flight English runs on: hosted API | 95.6%93.6%–97.0% | 16.7%answered 99% | Not measured | $1.427$1,427subscription, priced as the API | 16.916 per request1 in flight |
| GPT-6.1 Sol low · chat Closed · Workhorseother setting of GPT-6.1 SolLatency 10.3 sCost $1.219$1,219 per 1,000per 1M GPT-6.1 Sol low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 95.8% (95%: 93.8–97.2%), 547 kept reading questions Non-hallucination with “I don't know” 14.8% · answered 99.3% · error rate 4.2% Cost $1.219$1,219 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 10.3 s median per request, 1 in flight English runs on: hosted API | 95.8%93.8%–97.2% | 14.8%answered 99% | Not measured | $1.219$1,219subscription, priced as the API | 10.316 per request1 in flight | |
| GPT-6.1 Sol low · chat · 1 per request Closed · Workhorseother setting of GPT-6.1 SolLatency 5.59 sCost $6.113$6,113 per 1,000per 1M GPT-6.1 Sol low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Accuracy in English 95.2% (95%: 93.1–96.7%), 547 kept reading questions Non-hallucination with “I don't know” 12.9% · answered 99.3% · error rate 5.0% Cost $6.113$6,113 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 5.59 s median per request, 1 in flight English runs on: hosted API | 95.2%93.1%–96.7% | 12.9%answered 99% | Not measured | $6.113$6,113subscription, priced as the API | 5.591 per request1 in flight | |
| 9 | Claude Opus 5.5 high · chat Closed · FlagshipLatency 11.3 sCost $4.519$4,519 per 1,000per 1M Claude Opus 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 94.5% (95%: 92.3–96.1%), 547 kept reading questions Non-hallucination with “I don't know” 6.5% · answered 99.6% · error rate 5.3% Cost $4.519$4,519 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 11.3 s median per request, 1 in flight English runs on: hosted API | 94.5%92.3%–96.1% | 6.5%answered 100% | Not measured | $4.519$4,519subscription, priced as the API | 11.316 per request1 in flight |
| Claude Opus 5.5 low · chat Closed · Flagshipother setting of Claude Opus 5.5Latency 7.64 sCost $1.087$1,087 per 1,000per 1M Claude Opus 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 93.1% (95%: 90.6–94.9%), 547 kept reading questions Non-hallucination with “I don't know” 4.7% · answered 99.6% · error rate 7.5% Cost $1.087$1,087 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.64 s median per request, 1 in flight English runs on: hosted API | 93.1%90.6%–94.9% | 4.7%answered 100% | Not measured | $1.087$1,087subscription, priced as the API | 7.6416 per request1 in flight | |
| Claude Opus 5.5 low · chat · 1 per request Closed · Flagshipother setting of Claude Opus 5.5Latency 3.50 sCost $10.094$10,094 per 1,000per 1M Claude Opus 5.5 low · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 94.7% (95%: 92.5–96.3%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 5.7% Cost $10.094$10,094 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 3.50 s median per request, 1 in flight English runs on: hosted API | 94.7%92.5%–96.3% | 0.0%answered 100% | Not measured | $10.094$10,094subscription, priced as the API | 3.501 per request1 in flight | |
| 10 | Qwen4-Max high · chat Closed · WorkhorseLatency 29.5 sCost $1.119$1,119 per 1,000per 1M Qwen4-Max high · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://openrouter.ai/qwen/qwen4-max, read 2026-11-16) Accuracy in English 94.5% (95%: 92.3–96.1%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 5.3% Cost $1.119$1,119 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 29.5 s median per request, 4 in flight English runs on: hosted API | 94.5%92.3%–96.1% | 0.0%answered 100% | Not measured | $1.119$1,119at list prices | 29.516 per request4 in flight |
| 11 | Gemini 4 Carbon high · chat Closed · FlagshipLatency 287.8 sCost $38.053$38,053 per 1,000per 1M Gemini 4 Carbon high · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Accuracy in English 94.3% (95%: 92.1–96.0%), 547 kept reading questions Non-hallucination with “I don't know” 25.0% · answered 99.3% · error rate 2.2% Cost $38.053$38,053 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 287.8 s median per request, 1 in flight English runs on: hosted API | 94.3%92.1%–96.0% | 25.0%answered 99% | Not measured | $38.053$38,053subscription, priced as the API | 287.816 per request1 in flight |
| 12 | KnowLine-31B-Gen1.5 Open · 31BLatency 1.12 sCost No list price KnowLine-31B-Gen1.5 · PelaAI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 93.6% (95%: 91.2–95.4%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 6.8% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0494, Decision Score (AA formula) 94.33 Latency: 1.12 s median per request, 1 in flight English runs on: DGX Spark | 93.6%91.2%–95.4% | 0.0%answered 100% | 0.0494Decision Score 94.33 | No list price | 1.121 per request1 in flight |
| 13 | Grok 4.7 high · chat Closed · WorkhorseLatency 197.2 sCost $6.851$6,851 per 1,000per 1M Grok 4.7 high · chat · xAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://docs.x.ai/docs/models/grok-4.7, read 2026-10-28) Accuracy in English 93.4% (95%: 91.0–95.2%), 547 kept reading questions Non-hallucination with “I don't know” 27.9% · answered 97.8% · error rate 5.8% Cost $6.851$6,851 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 197.2 s median per request, 4 in flight English runs on: hosted API | 93.4%91.0%–95.2% | 27.9%answered 98% | Not measured | $6.851$6,851subscription, priced as the API | 197.216 per request4 in flight |
| 14 | Copernicus-31B Open · 31BLatency 0.188 sCost $0.142$142 per 1,000per 1M Copernicus-31B · Poindexter Labs · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 92.9% (95%: 90.4–94.7%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 7.1% Cost $0.142$142 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0556, Decision Score (AA formula) 93.66 Latency: 0.188 s median per request, 1 in flight English runs on: RTX PRO 6000 | 92.9%90.4%–94.7% | 0.0%answered 100% | 0.0556Decision Score 93.66 | $0.142$142from GPU time | 0.1881 per request1 in flight |
| 15 | Claude Haiku 5.5 max · chat Closed · EfficientLatency 148.9 sCost $1.261$1,261 per 1,000per 1M Claude Haiku 5.5 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/anthropic/claude-haiku-5.5, read 2026-11-10) Accuracy in English 92.7% (95%: 90.2–94.6%), 547 kept reading questions Non-hallucination with “I don't know” 7.8% · answered 96.3% · error rate 8.9% Cost $1.261$1,261 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 148.9 s median per request, 4 in flight English runs on: hosted API | 92.7%90.2%–94.6% | 7.8%answered 96% | Not measured | $1.261$1,261at list prices | 148.916 per request4 in flight |
| 16 | DeepSeek-V4.1-Flash reasoning on · chat · 1017 Open · 552B MoE · 16B activeLatency 189.4 sCost $0.421$421 per 1,000per 1M DeepSeek-V4.1-Flash reasoning on · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy in English 92.5% (95%: 90.0–94.4%), 547 kept reading questions Non-hallucination with “I don't know” 7.1% · answered 99.5% · error rate 7.2% Cost $0.421$421 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 189.4 s median per request, 8 in flight English runs on: RTX PRO 6000 | 92.5%90.0%–94.4% | 7.1%answered 99% | Not measured | $0.421$421at the maker's list price | 189.416 per request8 in flight |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017 Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-FlashLatency 6.59 sCost $0.079$79.31 per 1,000per 1M DeepSeek-V4.1-Flash reasoning off · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Accuracy in English 87.0% (95%: 83.9–89.6%), 547 kept reading questions Non-hallucination with “I don't know” 5.4% · answered 99.3% · error rate 12.9% Cost $0.079$79.31 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 6.59 s median per request, 1 in flight English runs on: RTX PRO 6000 | 87.0%83.9%–89.6% | 5.4%answered 99% | Not measured | $0.079$79.31at the maker's list price | 6.5916 per request1 in flight | |
| 17 | REAX 1.3 Open · 57BLatency 0.229 sCost $0.169$169 per 1,000per 1M REAX 1.3 · REAX · decision model Two models run together, option probabilities averaged 50/50: 57.36B in all, all used on every question. Accuracy in English 92.5% (95%: 90.0–94.4%), 547 kept reading questions Non-hallucination with “I don't know” 6.7% · answered 99.5% · error rate 7.7% Cost $0.169$169 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0500, Decision Score (AA formula) 93.75 Latency: 0.229 s median per request, 1 in flight English runs on: RTX PRO 6000 | 92.5%90.0%–94.4% | 6.7%answered 99% | 0.0500Decision Score 93.75 | $0.169$169from GPU time | 0.2291 per request1 in flight |
| 18 | Kimi K3.1 Pro low · chat Open · 2.8T MoE · 104B activeLatency 16.6 sCost $0.984$984 per 1,000per 1M Kimi K3.1 Pro low · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Accuracy in English 92.3% (95%: 89.8–94.3%), 547 kept reading questions Non-hallucination with “I don't know” 7.3% · answered 99.5% · error rate 7.0% Cost $0.984$984 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 16.6 s median per request, 4 in flight English runs on: hosted API | 92.3%89.8%–94.3% | 7.3%answered 99% | Not measured | $0.984$984at list prices | 16.616 per request4 in flight |
| Kimi K3.1 Pro no reasoning · chat Open · 2.8T MoE · 104B activeother setting of Kimi K3.1 ProLatency 6.31 sCost $0.735$735 per 1,000per 1M Kimi K3.1 Pro no reasoning · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Accuracy in English 91.4% (95%: 88.8–93.5%), 547 kept reading questions Non-hallucination with “I don't know” 7.5% · answered 99.3% · error rate 9.0% Cost $0.735$735 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 6.31 s median per request, 4 in flight English runs on: hosted API | 91.4%88.8%–93.5% | 7.5%answered 99% | Not measured | $0.735$735at list prices | 6.3116 per request4 in flight | |
| 19 | KnowLine-31B-Gen2 Open · 31BLatency 0.171 sCost $0.127$127 per 1,000per 1M KnowLine-31B-Gen2 · PelaAI · decision model Base model: Gemma 4 31B Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 92.3% (95%: 89.8–94.3%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 7.3% Cost $0.127$127 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0558, Decision Score (AA formula) 93.37 Latency: 0.171 s median per request, 1 in flight English runs on: RTX PRO 6000 | 92.3%89.8%–94.3% | 0.0%answered 100% | 0.0558Decision Score 93.37 | $0.127$127from GPU time | 0.1711 per request1 in flight |
| 20 | Karpo Decision 31B Open · 31BLatency 0.207 sCost $0.145$145 per 1,000per 1M Karpo Decision 31B · MachinePulse · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 92.1% (95%: 89.6–94.1%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 7.7% Cost $0.145$145 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0568, Decision Score (AA formula) 93.23 Latency: 0.207 s median per request, 1 in flight English runs on: RTX PRO 6000 | 92.1%89.6%–94.1% | 0.0%answered 100% | 0.0568Decision Score 93.23 | $0.145$145from GPU time | 0.2071 per request1 in flight |
| 21 | Qwen3.5-397B-A17B high · chat Open · 397B MoE · 17B activeLatency 162.9 sCost $1.742$1,742 per 1,000per 1M Qwen3.5-397B-A17B high · chat · Alibaba · chat model Mixture of experts: 397B total, about 17B active per step. Size class uses the total (397B). Accuracy in English 91.4% (95%: 88.8–93.5%), 547 kept reading questions Non-hallucination with “I don't know” 6.2% · answered 99.5% · error rate 8.3% Cost $1.742$1,742 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 162.9 s median per request, 4 in flight English runs on: hosted API | 91.4%88.8%–93.5% | 6.2%answered 99% | Not measured | $1.742$1,742at list prices | 162.916 per request4 in flight |
| 22 | Gemini Decision 1 20261111 Closed · EfficientLatency 0.235 sCost $0.040$39.51 per 1,000per 1M Gemini Decision 1 20261111 · Google · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/google/gemini-decision-1, read 2026-11-11) Accuracy in English 91.0% (95%: 88.4–93.2%), 547 kept reading questions Non-hallucination with “I don't know” 9.1% · answered 99.1% · error rate 9.2% Cost $0.040$39.51 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0642, Decision Score (AA formula) 92.31 Latency: 0.235 s median per request, 1 in flight English runs on: hosted API | 91.0%88.4%–93.2% | 9.1%answered 99% | 0.0642Decision Score 92.31 | $0.040$39.51at list prices | 0.2351 per request1 in flight |
| 23 | Logits Sigma-1 Duo Open · 57BLatency 0.263 sCost $0.191$191 per 1,000per 1M Logits Sigma-1 Duo · Logits · decision model Parameters: 57.36B dense, as the maker states it. Accuracy in English 91.0% (95%: 88.4–93.2%), 547 kept reading questions Non-hallucination with “I don't know” 4.3% · answered 99.6% · error rate 8.1% Cost $0.191$191 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0595, Decision Score (AA formula) 92.55 Latency: 0.263 s median per request, 1 in flight English runs on: RTX PRO 6000 | 91.0%88.4%–93.2% | 4.3%answered 100% | 0.0595Decision Score 92.55 | $0.191$191from GPU time | 0.2631 per request1 in flight |
| 24 | GLiDE 27B Open · 27BLatency 0.545 sCost No list price GLiDE 27B · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 90.9% (95%: 88.2–93.0%), 547 kept reading questions Non-hallucination with “I don't know” 37.9% · answered 96.0% · error rate 6.9% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0613, Decision Score (AA formula) 92.36 Latency: 0.545 s median per request, 1 in flight English runs on: DGX Spark | 90.9%88.2%–93.0% | 37.9%answered 96% | 0.0613Decision Score 92.36 | No list price | 0.5451 per request1 in flight |
| 25 | d1-ultra Closed · EfficientLatency 0.220 sCost $0.014$13.98 per 1,000per 1M d1-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-ultra, read 2026-10-16) Accuracy in English 90.7% (95%: 87.9–92.8%), 547 kept reading questions Non-hallucination with “I don't know” 56.7% · answered 90.7% · error rate 7.9% Cost $0.014$13.98 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0684, Decision Score (AA formula) 91.92 Latency: 0.220 s median per request, 1 in flight English runs on: hosted API | 90.7%87.9%–92.8% | 56.7%answered 91% | 0.0684Decision Score 91.92 | $0.014$13.98at list prices | 0.2201 per request1 in flight |
| 26 | Sieve 31B v3 Open · 31BLatency 1.06 sCost No list price Sieve 31B v3 · Sthānika AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 90.7% (95%: 87.9–92.8%), 547 kept reading questions Non-hallucination with “I don't know” 2.2% · answered 99.8% · error rate 8.2% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0565, Decision Score (AA formula) 92.51 Latency: 1.06 s median per request, 1 in flight English runs on: DGX Spark | 90.7%87.9%–92.8% | 2.2%answered 100% | 0.0565Decision Score 92.51 | No list price | 1.061 per request1 in flight |
| 27 | Turion Jev 31B Open · 31BLatency 0.445 sCost No list price Turion Jev 31B · Turion.AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 90.7% (95%: 87.9–92.8%), 547 kept reading questions Non-hallucination with “I don't know” 8.5% · answered 99.3% · error rate 7.9% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0729, Decision Score (AA formula) 91.69 Latency: 0.445 s median per request, 1 in flight English runs on: DGX Spark | 90.7%87.9%–92.8% | 8.5%answered 99% | 0.0729Decision Score 91.69 | No list price | 0.4451 per request1 in flight |
| 28 | Gemma 4 31B reasoning off · chat Open · 31BLatency 3.19 sCost $0.053$52.81 per 1,000per 1M Gemma 4 31B reasoning off · chat · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 90.5% (95%: 87.7–92.7%), 547 kept reading questions Non-hallucination with “I don't know” 3.6% · answered 99.6% · error rate 9.7% Cost $0.053$52.81 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 3.19 s median per request, 1 in flight English runs on: RTX PRO 6000 | 90.5%87.7%–92.7% | 3.6%answered 100% | Not measured | $0.053$52.81at a host's list price for the same weights | 3.1916 per request1 in flight |
| Gemma 4 31B reasoning off · chat · 1 per request Open · 31Bother setting of Gemma 4 31BLatency 0.299 sCost $0.070$70.17 per 1,000per 1M Gemma 4 31B reasoning off · chat · 1 per request · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Accuracy in English 89.9% (95%: 87.1–92.2%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 10.8% Cost $0.070$70.17 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Latency: 0.299 s median per request, 1 in flight English runs on: RTX PRO 6000 | 89.9%87.1%–92.2% | 0.0%answered 100% | Not measured | $0.070$70.17at a host's list price for the same weights | 0.2991 per request1 in flight | |
| – | Perplexity Decider 27B v1.2 API · with web Open · 27BLatency 0.484 sCost ≤ $1.036≤ $1,036 per 1,000per 1M Perplexity Decider 27B v1.2 API · with web · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 90.5% (95%: 87.7–92.7%), 547 kept reading questions Non-hallucination with “I don't know” 15.2% · answered 98.2% · error rate 10.4% Cost ≤ $1.036≤ $1,036 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0771, Decision Score (AA formula) 91.39 Latency: 0.484 s median per request, 1 in flight English runs on: hosted API | 90.5%87.7%–92.7% | 15.2%answered 98% | 0.0771Decision Score 91.39 | ≤ $1.036≤ $1,036at list prices | 0.4841 per request1 in flight |
| 29 | GLiDE v1.2 Open · 27BLatency 0.104 sCost $0.015$14.55 per 1,000per 1M GLiDE v1.2 · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 90.3% (95%: 87.5–92.5%), 547 kept reading questions Non-hallucination with “I don't know” 45.5% · answered 94.5% · error rate 7.0% Cost $0.015$14.55 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0639, Decision Score (AA formula) 91.96 Latency: 0.104 s median per request, 1 in flight English runs on: RTX PRO 6000 | 90.3%87.5%–92.5% | 45.5%answered 95% | 0.0639Decision Score 91.96 | $0.015$14.55at the maker's list price | 0.1041 per request1 in flight |
| 30 | d1.2-ultra Closed · EfficientLatency 0.221 sCost $0.014$13.98 per 1,000per 1M d1.2-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Accuracy in English 90.3% (95%: 87.5–92.5%), 547 kept reading questions Non-hallucination with “I don't know” 89.3% · answered 80.1% · error rate 3.0% Cost $0.014$13.98 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0664, Decision Score (AA formula) 91.84 Latency: 0.221 s median per request, 1 in flight English runs on: hosted API | 90.3%87.5%–92.5% | 89.3%answered 80% | 0.0664Decision Score 91.84 | $0.014$13.98at list prices | 0.2211 per request1 in flight |
| 31 | Mammoth-1 Open · 27BLatency 0.189 sCost $0.040$40.07 per 1,000per 1M Mammoth-1 · Base Labs · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Accuracy in English 90.1% (95%: 87.3–92.4%), 547 kept reading questions Non-hallucination with “I don't know” 50.7% · answered 93.2% · error rate 7.1% Cost $0.040$40.07 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0663, Decision Score (AA formula) 91.75 Latency: 0.189 s median per request, 1 in flight English runs on: RTX PRO 6000 | 90.1%87.3%–92.4% | 50.7%answered 93% | 0.0663Decision Score 91.75 | $0.040$40.07at the maker's list price | 0.1891 per request1 in flight |
| 32 | d3 Open · 26BLatency 0.074 sCost $0.055$55.35 per 1,000per 1M d3 · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 89.8% (95%: 86.9–92.0%), 547 kept reading questions Non-hallucination with “I don't know” 1.7% · answered 99.8% · error rate 10.6% Cost $0.055$55.35 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0646, Decision Score (AA formula) 91.65 Latency: 0.074 s median per request, 1 in flight English runs on: RTX PRO 6000 | 89.8%86.9%–92.0% | 1.7%answered 100% | 0.0646Decision Score 91.65 | $0.055$55.35from GPU time | 0.0741 per request1 in flight |
| 33 | GLM-5.3-Flash reasoning on · chat Open · 320B MoE · 18B activeLatency 33.1 sCost $0.081$81.38 per 1,000per 1M GLM-5.3-Flash reasoning on · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Accuracy in English 89.6% (95%: 86.7–91.9%), 547 kept reading questions Non-hallucination with “I don't know” 1.6% · answered 99.8% · error rate 11.2% Cost $0.081$81.38 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 33.1 s median per request, 8 in flight English runs on: RTX PRO 6000 | 89.6%86.7%–91.9% | 1.6%answered 100% | Not measured | $0.081$81.38at the maker's list price | 33.116 per request8 in flight |
| GLM-5.3-Flash reasoning off · chat Open · 320B MoE · 18B activeother setting of GLM-5.3-FlashLatency 3.37 sCost $0.046$46.07 per 1,000per 1M GLM-5.3-Flash reasoning off · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Accuracy in English 87.2% (95%: 84.1–89.7%), 547 kept reading questions Non-hallucination with “I don't know” 5.6% · answered 99.3% · error rate 12.3% Cost $0.046$46.07 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 3.37 s median per request, 1 in flight English runs on: RTX PRO 6000 | 87.2%84.1%–89.7% | 5.6%answered 99% | Not measured | $0.046$46.07at the maker's list price | 3.3716 per request1 in flight | |
| 34 | Drex Max Open · 27BLatency 0.081 sCost $0.024$23.56 per 1,000per 1M Drex Max · Nace.AI · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 89.6% (95%: 86.7–91.9%), 547 kept reading questions Non-hallucination with “I don't know” 3.2% · answered 99.6% · error rate 11.0% Cost $0.024$23.56 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0658, Decision Score (AA formula) 91.50 Latency: 0.081 s median per request, 1 in flight English runs on: RTX PRO 6000 | 89.6%86.7%–91.9% | 3.2%answered 100% | 0.0658Decision Score 91.50 | $0.024$23.56at the maker's list price | 0.0811 per request1 in flight |
| 35 | Muse Ember max · chat Closed · EfficientLatency 11.6 sCost $0.245$245 per 1,000per 1M Muse Ember max · chat · Meta · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/meta/muse-ember, read 2026-11-10) Accuracy in English 89.6% (95%: 86.7–91.9%), 547 kept reading questions Non-hallucination with “I don't know” 5.3% · answered 99.5% · error rate 9.9% Cost $0.245$245 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 11.6 s median per request, 4 in flight English runs on: hosted API | 89.6%86.7%–91.9% | 5.3%answered 99% | Not measured | $0.245$245at list prices | 11.616 per request4 in flight |
| 36 | Perplexity Decider 27B v1.3 Open · 27BLatency 0.115 sCost $0.083$82.63 per 1,000per 1M Perplexity Decider 27B v1.3 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 89.6% (95%: 86.7–91.9%), 547 kept reading questions Non-hallucination with “I don't know” 8.8% · answered 99.1% · error rate 9.6% Cost $0.083$82.63 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0714, Decision Score (AA formula) 91.22 Latency: 0.115 s median per request, 1 in flight English runs on: RTX PRO 6000 | 89.6%86.7%–91.9% | 8.8%answered 99% | 0.0714Decision Score 91.22 | $0.083$82.63from GPU time | 0.1151 per request1 in flight |
| 37 | d3-ultra Open · 26BLatency 0.149 sCost $0.106$106 per 1,000per 1M d3-ultra · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 89.4% (95%: 86.5–91.7%), 547 kept reading questions Non-hallucination with “I don't know” 8.2% · answered 99.1% · error rate 10.3% Cost $0.106$106 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0704, Decision Score (AA formula) 91.18 Latency: 0.149 s median per request, 1 in flight English runs on: RTX PRO 6000 | 89.4%86.5%–91.7% | 8.2%answered 99% | 0.0704Decision Score 91.18 | $0.106$106from GPU time | 0.1491 per request1 in flight |
| 38 | Qwen3.8-27B low · chat Open · 27BLatency 71.7 sCost $0.853$853 per 1,000per 1M Qwen3.8-27B low · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy in English 89.0% (95%: 86.1–91.4%), 547 kept reading questions Non-hallucination with “I don't know” 3.6% · answered 99.6% · error rate 9.7% Cost $0.853$853 per 1,000 questionsper 1M questions (at the maker's list price) · 16 questions per request Latency: 71.7 s median per request, 4 in flight English runs on: RTX PRO 6000 | 89.0%86.1%–91.4% | 3.6%answered 100% | Not measured | $0.853$853at the maker's list price | 71.716 per request4 in flight |
| Qwen3.8-27B yes/no scoring Open · 27Bother setting of Qwen3.8-27BLatency 0.413 sCost $0.741$741 per 1,000per 1M Qwen3.8-27B yes/no scoring · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy in English 84.8% (95%: 81.6–87.6%), 547 kept reading questions Cost $0.741$741 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.413 s median per request, 1 in flight English runs on: RTX PRO 6000 | 84.8%81.6%–87.6% | Not run | Not measured | $0.741$741at the maker's list price | 0.4131 per request1 in flight | |
| Qwen3.8-27B stated confidence · reasoning off · chat Open · 27Bother setting of Qwen3.8-27BLatency 0.390 sCost $0.212$212 per 1,000per 1M Qwen3.8-27B stated confidence · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy in English 84.8% (95%: 81.6–87.6%), 547 kept reading questions Non-hallucination with “I don't know” 3.7% · answered 99.5% · error rate 14.3% Cost $0.212$212 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.390 s median per request, 1 in flight English runs on: RTX PRO 6000 | 84.8%81.6%–87.6% | 3.7%answered 99% | Not measured | $0.212$212at the maker's list price | 0.3901 per request1 in flight | |
| Qwen3.8-27B vote share · reasoning off · chat Open · 27Bother setting of Qwen3.8-27BLatency 1.70 sCost $0.645$645 per 1,000per 1M Qwen3.8-27B vote share · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Accuracy in English 83.9% (95%: 80.6–86.8%), 547 kept reading questions Non-hallucination with “I don't know” 3.4% · answered 99.5% · error rate 15.4% Cost $0.645$645 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 1.70 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.9%80.6%–86.8% | 3.4%answered 99% | Not measured | $0.645$645at the maker's list price | 1.701 per request1 in flight | |
| 39 | MiMo-V2.6-Flash high · chat Open · 309B MoE · 15B activeLatency 6.67 sCost $0.298$298 per 1,000per 1M MiMo-V2.6-Flash high · chat · Xiaomi · chat model Mixture of experts: 309B total, about 15B active per step. Size class uses the total (309B). Accuracy in English 89.0% (95%: 86.1–91.4%), 547 kept reading questions Non-hallucination with “I don't know” 1.7% · answered 99.8% · error rate 10.6% Cost $0.298$298 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 6.67 s median per request, 8 in flight English runs on: hosted API | 89.0%86.1%–91.4% | 1.7%answered 100% | Not measured | $0.298$298at list prices | 6.671 per request8 in flight |
| 40 | d1.2 Closed · EfficientLatency 0.219 sCost $0.007$6.66 per 1,000per 1M d1.2 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Accuracy in English 89.0% (95%: 86.1–91.4%), 547 kept reading questions Non-hallucination with “I don't know” 61.5% · answered 89.2% · error rate 7.6% Cost $0.007$6.66 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0801, Decision Score (AA formula) 90.51 Latency: 0.219 s median per request, 1 in flight English runs on: hosted API | 89.0%86.1%–91.4% | 61.5%answered 89% | 0.0801Decision Score 90.51 | $0.007$6.66at list prices | 0.2191 per request1 in flight |
| 41 | Perplexity Decider 27B v1.2 DGX Spark Open · 27BLatency 0.126 sCost $0.008$7.85 per 1,000per 1M Perplexity Decider 27B v1.2 DGX Spark · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 88.7% (95%: 85.7–91.1%), 547 kept reading questions Non-hallucination with “I don't know” 3.0% · answered 99.6% · error rate 11.7% Cost $0.008$7.85 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0717, Decision Score (AA formula) 90.75 Latency: 0.126 s median per request, 1 in flight English runs on: RTX PRO 6000 | 88.7%85.7%–91.1% | 3.0%answered 100% | 0.0717Decision Score 90.75 | $0.008$7.85at the maker's list price | 0.1261 per request1 in flight |
| Perplexity Decider 27B v1.2 API Open · 27Bother setting of Perplexity Decider 27B v1.2Latency 0.256 sCost $0.008$7.79 per 1,000per 1M Perplexity Decider 27B v1.2 API · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 88.5% (95%: 85.5–90.9%), 547 kept reading questions Non-hallucination with “I don't know” 4.5% · answered 99.5% · error rate 11.6% Cost $0.008$7.79 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0721, Decision Score (AA formula) 90.64 Latency: 0.256 s median per request, 1 in flight English runs on: hosted API | 88.5%85.5%–90.9% | 4.5%answered 99% | 0.0721Decision Score 90.64 | $0.008$7.79at list prices | 0.2561 per request1 in flight | |
| 42 | Perplexity Decider 27B v1 Open · 27BLatency 0.247 sCost $0.016$15.59 per 1,000per 1M Perplexity Decider 27B v1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Accuracy in English 88.5% (95%: 85.5–90.9%), 547 kept reading questions Cost $0.016$15.59 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0720, Decision Score (AA formula) 90.64 Latency: 0.247 s median per request, 1 in flight English runs on: hosted API | 88.5%85.5%–90.9% | Not run | 0.0720Decision Score 90.64 | $0.016$15.59at list prices | 0.2471 per request1 in flight |
| 43 | Drex Max v1.1 Closed · Efficient (provisional)Latency 0.335 sCost $0.023$23.38 per 1,000per 1M Drex Max v1.1 · Nace.AI · decision model Parameters: not disclosed by maker Class: Efficient (provisional), by its introductory price, the only one published: $0.03 per 1M input tokens (https://nace.ai/developers, read 2026-11-10) Accuracy in English 88.5% (95%: 85.5–90.9%), 547 kept reading questions Non-hallucination with “I don't know” 51.2% · answered 92.1% · error rate 8.1% Cost $0.023$23.38 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0685, Decision Score (AA formula) 90.81 Latency: 0.335 s median per request, 1 in flight English runs on: hosted API | 88.5%85.5%–90.9% | 51.2%answered 92% | 0.0685Decision Score 90.81 | $0.023$23.38at list prices | 0.3351 per request1 in flight |
| 44 | d1.1 Closed · EfficientLatency 0.173 sCost $0.007$6.66 per 1,000per 1M d1.1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-24) Accuracy in English 88.3% (95%: 85.3–90.7%), 547 kept reading questions Cost $0.007$6.66 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0802, Decision Score (AA formula) 90.14 Latency: 0.173 s median per request, 1 in flight English runs on: hosted API | 88.3%85.3%–90.7% | Not run | 0.0802Decision Score 90.14 | $0.007$6.66at list prices | 0.1731 per request1 in flight |
| 45 | Inkling-Small high · chat · Jul 2026 Open · 276B MoE · 12B activeLatency 25.3 sCost $0.425$425 per 1,000per 1M Inkling-Small high · chat · Jul 2026 · Thinking Machines · chat model Mixture of experts: 276B total (counted from the released files: 265.96B), about 12B active per step. Size class uses the total (276B). Accuracy in English 88.3% (95%: 85.3–90.7%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 11.0% Cost $0.425$425 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 25.3 s median per request, 4 in flight English runs on: hosted API | 88.3%85.3%–90.7% | 0.0%answered 100% | Not measured | $0.425$425at list prices | 25.316 per request4 in flight |
| 46 | GLM-5.3 low · chat Open · WorkhorseLatency 6.84 sCost $0.408$408 per 1,000per 1M GLM-5.3 low · chat · Z.ai · chat model Parameters: not stated by the maker (a mixture of experts); counted from the released files: 753.33B. Class: Workhorse, by its standard list price: $1.40 per 1M input tokens (https://openrouter.ai/z-ai/glm-5.3, read 2026-10-28) Accuracy in English 87.8% (95%: 84.7–90.2%), 547 kept reading questions Non-hallucination with “I don't know” 3.4% · answered 99.6% · error rate 10.3% Cost $0.408$408 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 6.84 s median per request, 4 in flight English runs on: hosted API | 87.8%84.7%–90.2% | 3.4%answered 100% | Not measured | $0.408$408at list prices | 6.8416 per request4 in flight |
| 47 | Jeffan-9B-1211 Open · 9.0BLatency 0.067 sCost $0.046$45.85 per 1,000per 1M Jeffan-9B-1211 · tattrongvu · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Accuracy in English 87.6% (95%: 84.5–90.1%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 13.0% Cost $0.046$45.85 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0946, Decision Score (AA formula) 89.05 Latency: 0.067 s median per request, 1 in flight English runs on: RTX PRO 6000 | 87.6%84.5%–90.1% | 0.0%answered 100% | 0.0946Decision Score 89.05 | $0.046$45.85from GPU time | 0.0671 per request1 in flight |
| 48 | Qwen3.8 Flash no reasoning · chat Closed · EfficientLatency 10.1 sCost $0.050$49.69 per 1,000per 1M Qwen3.8 Flash no reasoning · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.15 per 1M input tokens (https://openrouter.ai/qwen/qwen3.8-flash, read 2026-11-11) Accuracy in English 87.0% (95%: 83.9–89.6%), 547 kept reading questions Non-hallucination with “I don't know” 5.6% · answered 99.3% · error rate 12.3% Cost $0.050$49.69 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 10.1 s median per request, 4 in flight English runs on: hosted API | 87.0%83.9%–89.6% | 5.6%answered 99% | Not measured | $0.050$49.69at list prices | 10.116 per request4 in flight |
| 49 | Qwen3.5-4B reasoning on · chat Open · 4.0BLatency 58.3 sCost $1.390$1,390 per 1,000per 1M Qwen3.5-4B reasoning on · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy in English 87.0% (95%: 83.9–89.6%), 547 kept reading questions Non-hallucination with “I don't know” 2.4% · answered 99.6% · error rate 15.2% Cost $1.390$1,390 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: not measured Latency: 58.3 s median per request, 32 in flight English runs on: RTX PRO 6000 | 87.0%83.9%–89.6% | 2.4%answered 100% | Not measureda plain answer line, no confidence | $1.390$1,390from GPU time | 58.31 per request32 in flight |
| 50 | d1 Closed · EfficientLatency 0.180 sCost $0.007$6.66 per 1,000per 1M d1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-18) Accuracy in English 86.8% (95%: 83.7–89.4%), 547 kept reading questions Non-hallucination with “I don't know” 2.9% · answered 99.6% · error rate 12.3% Cost $0.007$6.66 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0832, Decision Score (AA formula) 89.26 Latency: 0.180 s median per request, 1 in flight English runs on: hosted API | 86.8%83.7%–89.4% | 2.9%answered 100% | 0.0832Decision Score 89.26 | $0.007$6.66at list prices | 0.1801 per request1 in flight |
| 51 | Gemma 4 12B Unified default reasoning · chat Open · 12BLatency 15.8 sCost $0.154$154 per 1,000per 1M Gemma 4 12B Unified default reasoning · chat · Google · chat model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Accuracy in English 86.8% (95%: 83.7–89.4%), 547 kept reading questions Non-hallucination with “I don't know” 5.5% · answered 98.9% · error rate 12.8% Cost $0.154$154 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-12B-it@43c9d01), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0; output cap 8,192 Confidence: not measured Latency: 15.8 s median per request, 32 in flight English runs on: RTX PRO 6000 | 86.8%83.7%–89.4% | 5.5%answered 99% | Not measureda plain answer line, no confidence | $0.154$154at a host's list price for the same weights | 15.81 per request32 in flight |
| 52 | StartLux-Decision-v1.1-9B-Instant Open · 9.0BLatency 0.156 sCost No list price StartLux-Decision-v1.1-9B-Instant · StartLux · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy in English 86.8% (95%: 83.7–89.4%), 547 kept reading questions Non-hallucination with “I don't know” 2.6% · answered 99.6% · error rate 13.6% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0965, Decision Score (AA formula) 88.59 Latency: 0.156 s median per request, 1 in flight English runs on: DGX Spark | 86.8%83.7%–89.4% | 2.6%answered 100% | 0.0965Decision Score 88.59 | No list price | 0.1561 per request1 in flight |
| 53 | Clef Open · 27BLatency 0.220 sCost $0.111$111 per 1,000per 1M Clef · Cloudflare · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.48B). Accuracy in English 86.7% (95%: 83.5–89.3%), 547 kept reading questions Non-hallucination with “I don't know” 1.3% · answered 99.8% · error rate 13.7% Cost $0.111$111 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0828, Decision Score (AA formula) 89.19 Latency: 0.220 s median per request, 1 in flight English runs on: RTX PRO 6000 | 86.7%83.5%–89.3% | 1.3%answered 100% | 0.0828Decision Score 89.19 | $0.111$111at the maker's list price | 0.2201 per request1 in flight |
| 54 | KnowLine-12B-Gen1 Open · 12BLatency 0.078 sCost $0.057$56.86 per 1,000per 1M KnowLine-12B-Gen1 · PelaAI · decision model Base model: Gemma 4 12B Unified Parameters: 12B dense, as the maker states it (counted from the released files: 11.96B). Accuracy in English 86.7% (95%: 83.5–89.3%), 547 kept reading questions Non-hallucination with “I don't know” 1.3% · answered 99.8% · error rate 13.9% Cost $0.057$56.86 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0933, Decision Score (AA formula) 88.66 Latency: 0.078 s median per request, 1 in flight English runs on: RTX PRO 6000 | 86.7%83.5%–89.3% | 1.3%answered 100% | 0.0933Decision Score 88.66 | $0.057$56.86from GPU time | 0.0781 per request1 in flight |
| 55 | GPT-6 Luna low · chat Closed · EfficientLatency 7.72 sCost $0.058$57.70 per 1,000per 1M GPT-6 Luna low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Accuracy in English 86.3% (95%: 83.2–88.9%), 547 kept reading questions Non-hallucination with “I don't know” 6.8% · answered 99.1% · error rate 12.7% Cost $0.058$57.70 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 7.72 s median per request, 1 in flight English runs on: hosted API | 86.3%83.2%–88.9% | 6.8%answered 99% | Not measured | $0.058$57.70subscription, priced as the API | 7.7216 per request1 in flight |
| 56 | OneJev-27B Open · 27BLatency 0.130 sCost $0.106$106 per 1,000per 1M OneJev-27B · OmniJev · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Accuracy in English 86.3% (95%: 83.2–88.9%), 547 kept reading questions Non-hallucination with “I don't know” 4.0% · answered 99.5% · error rate 13.2% Cost $0.106$106 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0927, Decision Score (AA formula) 88.51 Latency: 0.130 s median per request, 1 in flight English runs on: RTX PRO 6000 | 86.3%83.2%–88.9% | 4.0%answered 99% | 0.0927Decision Score 88.51 | $0.106$106from GPU time | 0.1301 per request1 in flight |
| 57 | GLiDE 9B Open · 7.9BLatency 0.207 sCost No list price GLiDE 9B · Fastino · decision model Parameters: 7.94B dense, as the maker states it (counted from the released files: 7.94B). Accuracy in English 86.3% (95%: 83.2–88.9%), 547 kept reading questions Non-hallucination with “I don't know” 16.7% · answered 97.4% · error rate 13.1% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0909, Decision Score (AA formula) 88.60 Latency: 0.207 s median per request, 1 in flight English runs on: DGX Spark | 86.3%83.2%–88.9% | 16.7%answered 97% | 0.0909Decision Score 88.60 | No list price | 0.2071 per request1 in flight |
| 58 | Claude Sonnet 5.5 high · chat Closed · WorkhorseLatency 3.85 sCost $1.998$1,998 per 1,000per 1M Claude Sonnet 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 86.1% (95%: 83.0–88.8%), 547 kept reading questions Non-hallucination with “I don't know” 4.8% · answered 99.5% · error rate 10.8% Cost $1.998$1,998 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 3.85 s median per request, 1 in flight English runs on: hosted API | 86.1%83.0%–88.8% | 4.8%answered 99% | Not measured | $1.998$1,998subscription, priced as the API | 3.8516 per request1 in flight |
| Claude Sonnet 5.5 high · chat · 1 per request Closed · Workhorseother setting of Claude Sonnet 5.5Latency 3.02 sCost $5.513$5,513 per 1,000per 1M Claude Sonnet 5.5 high · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 91.6% (95%: 89.0–93.6%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 6.9% Cost $5.513$5,513 per 1,000 questionsper 1M questions (subscription, priced as the API) · 1 question per request Latency: 3.02 s median per request, 1 in flight English runs on: hosted API | 91.6%89.0%–93.6% | 0.0%answered 100% | Not measured | $5.513$5,513subscription, priced as the API | 3.021 per request1 in flight | |
| Claude Sonnet 5.5 low · chat Closed · Workhorseother setting of Claude Sonnet 5.5Latency 3.68 sCost $0.354$354 per 1,000per 1M Claude Sonnet 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Accuracy in English 83.9% (95%: 80.6–86.8%), 547 kept reading questions Non-hallucination with “I don't know” 4.4% · answered 99.3% · error rate 15.8% Cost $0.354$354 per 1,000 questionsper 1M questions (subscription, priced as the API) · 16 questions per request Latency: 3.68 s median per request, 1 in flight English runs on: hosted API | 83.9%80.6%–86.8% | 4.4%answered 99% | Not measured | $0.354$354subscription, priced as the API | 3.6816 per request1 in flight | |
| 59 | Jev 1.14 Closed · EfficientLatency 0.181 sCost $0.027$26.64 per 1,000per 1M Jev 1.14 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://docs.typesafe.ai/models, read 2026-10-16) Accuracy in English 86.1% (95%: 83.0–88.8%), 547 kept reading questions Non-hallucination with “I don't know” 4.0% · answered 99.5% · error rate 13.2% Cost $0.027$26.64 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0778, Decision Score (AA formula) 89.16 Latency: 0.181 s median per request, 1 in flight English runs on: hosted API | 86.1%83.0%–88.8% | 4.0%answered 99% | 0.0778Decision Score 89.16 | $0.027$26.64at list prices | 0.1811 per request1 in flight |
| 60 | Jev 1.13 Closed · EfficientLatency 0.173 sCost $0.027$26.64 per 1,000per 1M Jev 1.13 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://typesafe.ai/blog/introducing-system-one-models-and-jev, read 2026-10-08) Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Cost $0.027$26.64 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0821, Decision Score (AA formula) 88.86 Latency: 0.173 s median per request, 1 in flight English runs on: hosted API | 85.9%82.8%–88.6% | Not run | 0.0821Decision Score 88.86 | $0.027$26.64at list prices | 0.1731 per request1 in flight |
| 61 | SubQ Snap Preview Closed · EfficientLatency 0.381 sCost $0.019$19.48 per 1,000per 1M SubQ Snap Preview · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.05 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-10-13) Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Cost $0.019$19.48 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0852, Decision Score (AA formula) 88.70 Latency: 0.381 s median per request, 1 in flight English runs on: hosted API | 85.9%82.8%–88.6% | Not run | 0.0852Decision Score 88.70 | $0.019$19.48at list prices | 0.3811 per request1 in flight |
| 62 | Drex 2.0 Open · 35B MoE · 3.0B activeLatency 0.283 sCost $0.019$19.42 per 1,000per 1M Drex 2.0 · Nace.AI · decision model Mixture of experts: 35B total (counted from the released files: 35.11B), about 3B active per step. Size class uses the total (35B). Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Cost $0.019$19.42 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0963, Decision Score (AA formula) 88.15 Latency: 0.283 s median per request, 1 in flight English runs on: hosted API | 85.9%82.8%–88.6% | Not run | 0.0963Decision Score 88.15 | $0.019$19.42at list prices | 0.2831 per request1 in flight |
| 63 | SubQ Snap 1.1 Closed · EfficientLatency 0.381 sCost $0.016$15.59 per 1,000per 1M SubQ Snap 1.1 · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-11-01) Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Non-hallucination with “I don't know” 25.0% · answered 96.2% · error rate 12.0% Cost $0.016$15.59 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0852, Decision Score (AA formula) 88.70 Latency: 0.381 s median per request, 1 in flight English runs on: hosted API | 85.9%82.8%–88.6% | 25.0%answered 96% | 0.0852Decision Score 88.70 | $0.016$15.59at list prices | 0.3811 per request1 in flight |
| 64 | KnowLine-9B-Gen1 Open · 9.0BLatency 0.059 sCost $0.042$41.55 per 1,000per 1M KnowLine-9B-Gen1 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Non-hallucination with “I don't know” 1.3% · answered 99.8% · error rate 13.9% Cost $0.042$41.55 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0928, Decision Score (AA formula) 88.32 Latency: 0.059 s median per request, 1 in flight English runs on: RTX PRO 6000 | 85.9%82.8%–88.6% | 1.3%answered 100% | 0.0928Decision Score 88.32 | $0.042$41.55from GPU time | 0.0591 per request1 in flight |
| 65 | Cerno 9B v1.2 Open · 9.0BLatency 0.211 sCost No list price Cerno 9B v1.2 · Apertix AI · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Accuracy in English 85.9% (95%: 82.8–88.6%), 547 kept reading questions Non-hallucination with “I don't know” 2.5% · answered 99.6% · error rate 14.1% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.0907, Decision Score (AA formula) 88.42 Latency: 0.211 s median per request, 1 in flight English runs on: DGX Spark | 85.9%82.8%–88.6% | 2.5%answered 100% | 0.0907Decision Score 88.42 | No list price | 0.2111 per request1 in flight |
| 66 | Mercury Decide Closed · EfficientLatency 0.310 sCost $0.008$7.79 per 1,000per 1M Mercury Decide · Inception · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://www.inceptionlabs.ai/, read 2026-11-10) Accuracy in English 85.7% (95%: 82.6–88.4%), 547 kept reading questions Non-hallucination with “I don't know” 13.4% · answered 98.0% · error rate 13.2% Cost $0.008$7.79 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0943, Decision Score (AA formula) 88.15 Latency: 0.310 s median per request, 1 in flight English runs on: hosted API | 85.7%82.6%–88.4% | 13.4%answered 98% | 0.0943Decision Score 88.15 | $0.008$7.79at list prices | 0.3101 per request1 in flight |
| 67 | Microsoft-Decision-1 Closed · EfficientLatency 0.315 sCost $0.014$14.01 per 1,000per 1M Microsoft-Decision-1 · Microsoft · decision model Base model: Qwen3.5-9B Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://openrouter.ai/microsoft/microsoft-decision-1, read 2026-10-17) Accuracy in English 85.7% (95%: 82.6–88.4%), 547 kept reading questions Non-hallucination with “I don't know” 3.8% · answered 99.5% · error rate 14.2% Cost $0.014$14.01 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0954, Decision Score (AA formula) 88.10 Latency: 0.315 s median per request, 1 in flight English runs on: hosted API | 85.7%82.6%–88.4% | 3.8%answered 99% | 0.0954Decision Score 88.10 | $0.014$14.01at list prices | 0.3151 per request1 in flight |
| 68 | Adjudex 9B v5 Open · 9.0BLatency 0.034 sCost $0.025$24.93 per 1,000per 1M Adjudex 9B v5 · newsbang · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Accuracy in English 85.7% (95%: 82.6–88.4%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 14.1% Cost $0.025$24.93 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0905, Decision Score (AA formula) 88.35 Latency: 0.034 s median per request, 1 in flight English runs on: RTX PRO 6000 | 85.7%82.6%–88.4% | 0.0%answered 100% | 0.0905Decision Score 88.35 | $0.025$24.93from GPU time | 0.0341 per request1 in flight |
| 69 | Open-Jev-27B-v1.1 Open · 27BLatency 0.495 sCost $0.351$351 per 1,000per 1M Open-Jev-27B-v1.1 · Zefan Cai · decision model Base model: Qwen3.8-27B + adapter Parameters: 27B dense, as the maker states it (counted from the released files: 27.80B). Accuracy in English 85.4% (95%: 82.2–88.1%), 547 kept reading questions Non-hallucination with “I don't know” 17.4% · answered 97.1% · error rate 14.3% Cost $0.351$351 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.3102, Decision Score (AA formula) 77.18 Latency: 0.495 s median per request, 1 in flight English runs on: RTX PRO 6000 | 85.4%82.2%–88.1% | 17.4%answered 97% | 0.3102Decision Score 77.18 | $0.351$351from GPU time | 0.4951 per request1 in flight |
| 70 | Elephant-1 Open · 9.0BLatency 0.070 sCost $0.014$14.02 per 1,000per 1M Elephant-1 · Base Labs · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.41B). Accuracy in English 85.4% (95%: 82.2–88.1%), 547 kept reading questions Non-hallucination with “I don't know” 25.6% · answered 95.8% · error rate 12.8% Cost $0.014$14.02 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0939, Decision Score (AA formula) 87.99 Latency: 0.070 s median per request, 1 in flight English runs on: RTX PRO 6000 | 85.4%82.2%–88.1% | 25.6%answered 96% | 0.0939Decision Score 87.99 | $0.014$14.02at the maker's list price | 0.0701 per request1 in flight |
| 71 | KnowLine-9B-Gen2 Open · 9.0BLatency 0.056 sCost $0.039$39.48 per 1,000per 1M KnowLine-9B-Gen2 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Accuracy in English 85.2% (95%: 82.0–87.9%), 547 kept reading questions Non-hallucination with “I don't know” 1.3% · answered 99.8% · error rate 14.3% Cost $0.039$39.48 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1044, Decision Score (AA formula) 87.38 Latency: 0.056 s median per request, 1 in flight English runs on: RTX PRO 6000 | 85.2%82.0%–87.9% | 1.3%answered 100% | 0.1044Decision Score 87.38 | $0.039$39.48from GPU time | 0.0561 per request1 in flight |
| 72 | MiniMax-M3.1-Flash high · chat Open · 428B MoE · 23B activeLatency 26.7 sCost $0.193$193 per 1,000per 1M MiniMax-M3.1-Flash high · chat · MiniMax · chat model Mixture of experts: 428B total, about 23B active per step. Size class uses the total (428B). Accuracy in English 84.5% (95%: 81.2–87.3%), 547 kept reading questions Non-hallucination with “I don't know” 3.4% · answered 99.5% · error rate 15.6% Cost $0.193$193 per 1,000 questionsper 1M questions (at list prices) · 16 questions per request Latency: 26.7 s median per request, 4 in flight English runs on: hosted API | 84.5%81.2%–87.3% | 3.4%answered 99% | Not measured | $0.193$193at list prices | 26.716 per request4 in flight |
| 73 | GPT-6 Luna Decisions API Closed · EfficientLatency 0.323 sCost $0.041$41.09 per 1,000per 1M GPT-6 Luna Decisions API · OpenAI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Accuracy in English 84.1% (95%: 80.8–86.9%), 547 kept reading questions Non-hallucination with “I don't know” 9.4% · answered 98.2% · error rate 17.9% Cost $0.041$41.09 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Latency: 0.323 s median per request, 4 in flight English runs on: hosted API | 84.1%80.8%–86.9% | 9.4%answered 98% | Unavailable | $0.041$41.09at list prices | 0.3231 per request4 in flight |
| 74 | celeris-1-decision Closed · EfficientLatency 0.390 sCost $0.016$16.09 per 1,000per 1M celeris-1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-10-10) Accuracy in English 84.1% (95%: 80.8–86.9%), 547 kept reading questions Non-hallucination with “I don't know” 71.3% · answered 83.2% · error rate 8.1% Cost $0.016$16.09 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.0952, Decision Score (AA formula) 87.29 Latency: 0.390 s median per request, 1 in flight English runs on: hosted API | 84.1%80.8%–86.9% | 71.3%answered 83% | 0.0952Decision Score 87.29 | $0.016$16.09at list prices | 0.3901 per request1 in flight |
| 75 | Cygnet Open · 12BLatency 0.043 sCost $0.041$41.42 per 1,000per 1M Cygnet · Blockbrain Labs · decision model Base model: Gemma 4 12B IT Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Accuracy in English 83.9% (95%: 80.6–86.8%), 547 kept reading questions Non-hallucination with “I don't know” 11.2% · answered 98.2% · error rate 14.7% Cost $0.041$41.42 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 1 question per request Confidence: chosen-answer Brier 0.0984, Decision Score (AA formula) 87.03 Latency: 0.043 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.9%80.6%–86.8% | 11.2%answered 98% | 0.0984Decision Score 87.03 | $0.041$41.42at a host's list price for the same weights | 0.0431 per request1 in flight |
| 76 | Winnow-12B Open · 12BLatency 0.070 sCost $0.050$49.59 per 1,000per 1M Winnow-12B · EldanRing · decision model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.91B). Accuracy in English 83.9% (95%: 80.6–86.8%), 547 kept reading questions Non-hallucination with “I don't know” 15.0% · answered 97.3% · error rate 16.0% Cost $0.050$49.59 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1156, Decision Score (AA formula) 86.18 Latency: 0.070 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.9%80.6%–86.8% | 15.0%answered 97% | 0.1156Decision Score 86.18 | $0.050$49.59from GPU time | 0.0701 per request1 in flight |
| 77 | celeris-1.1-decision Closed · EfficientLatency 0.385 sCost $0.002$1.61 per 1,000per 1M celeris-1.1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.004 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-11-09) Accuracy in English 83.7% (95%: 80.4–86.6%), 547 kept reading questions Non-hallucination with “I don't know” 74.8% · answered 82.1% · error rate 7.3% Cost $0.002$1.61 per 1,000 questionsper 1M questions (at list prices) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: Celeris System One API Confidence: chosen-answer Brier 0.0946, Decision Score (AA formula) 87.14 Latency: 0.385 s median per request, 1 in flight English runs on: hosted API | 83.7%80.4%–86.6% | 74.8%answered 82% | 0.0946Decision Score 87.14 | $0.002$1.61at list prices | 0.3851 per request1 in flight |
| 78 | Perplexity Decider Flash v1.3 Open · 4.0BLatency 0.033 sCost $0.023$23.17 per 1,000per 1M Perplexity Decider Flash v1.3 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Accuracy in English 83.7% (95%: 80.4–86.6%), 547 kept reading questions Non-hallucination with “I don't know” 7.2% · answered 98.7% · error rate 16.7% Cost $0.023$23.17 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0980, Decision Score (AA formula) 86.96 Latency: 0.033 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.7%80.4%–86.6% | 7.2%answered 99% | 0.0980Decision Score 86.96 | $0.023$23.17from GPU time | 0.0331 per request1 in flight |
| 79 | Valen-Decision-1.5-9B Open · 9.0BLatency 0.059 sCost $0.041$40.55 per 1,000per 1M Valen-Decision-1.5-9B · Valen-Team · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.40B). Accuracy in English 83.7% (95%: 80.4–86.6%), 547 kept reading questions Non-hallucination with “I don't know” 1.2% · answered 99.8% · error rate 15.4% Cost $0.041$40.55 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1087, Decision Score (AA formula) 86.43 Latency: 0.059 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.7%80.4%–86.6% | 1.2%answered 100% | 0.1087Decision Score 86.43 | $0.041$40.55from GPU time | 0.0591 per request1 in flight |
| 80 | Dexy-Decision-9B-V2 Open · 9.0BLatency 0.036 sCost $0.025$25.41 per 1,000per 1M Dexy-Decision-9B-V2 · Patrick-18 · decision model Base model: vllm-sr/d3-flash Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Accuracy in English 83.5% (95%: 80.2–86.4%), 547 kept reading questions Non-hallucination with “I don't know” 1.1% · answered 99.8% · error rate 16.5% Cost $0.025$25.41 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.0953, Decision Score (AA formula) 87.01 Latency: 0.036 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.5%80.2%–86.4% | 1.1%answered 100% | 0.0953Decision Score 87.01 | $0.025$25.41from GPU time | 0.0361 per request1 in flight |
| 81 | vera-medium-omni Open · 4.2BLatency 0.054 sCost $0.037$36.63 per 1,000per 1M vera-medium-omni · scar-ai · decision model Base model: Qwen3.5-4B Parameters: 4.2B dense, as the maker states it (counted from the released files: 4.21B). Accuracy in English 83.5% (95%: 80.2–86.4%), 547 kept reading questions Non-hallucination with “I don't know” 7.4% · answered 98.7% · error rate 16.1% Cost $0.037$36.63 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1086, Decision Score (AA formula) 86.34 Latency: 0.054 s median per request, 1 in flight English runs on: RTX PRO 6000 | 83.5%80.2%–86.4% | 7.4%answered 99% | 0.1086Decision Score 86.34 | $0.037$36.63from GPU time | 0.0541 per request1 in flight |
| 82 | StartLux-Decision-v1.1-4B-Instant Open · 4.0BLatency 0.075 sCost No list price StartLux-Decision-v1.1-4B-Instant · StartLux · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy in English 82.8% (95%: 79.4–85.7%), 547 kept reading questions Non-hallucination with “I don't know” 0.0% · answered 100.0% · error rate 17.2% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1011, Decision Score (AA formula) 86.35 Latency: 0.075 s median per request, 1 in flight English runs on: DGX Spark | 82.8%79.4%–85.7% | 0.0%answered 100% | 0.1011Decision Score 86.35 | No list price | 0.0751 per request1 in flight |
| 83 | LFM2.5-24B-A2B default reasoning · chat Open · 24B MoE · 2.5B activeLatency 74.2 sCost $1.764$1,764 per 1,000per 1M LFM2.5-24B-A2B default reasoning · chat · Liquid AI · chat model Mixture of experts: 24B total (counted from the released files: 23.97B), about 2.5B active per step. Size class uses the total (24B). Accuracy in English 82.6% (95%: 79.2–85.6%), 547 kept reading questions Non-hallucination with “I don't know” 13.2% · answered 97.4% · error rate 17.3% Cost $1.764$1,764 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-24B-A2B@1769516), one question per request, plain answer line ANSWER: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.1, top_k 50; output cap 32,768 Confidence: not measured Latency: 74.2 s median per request, 32 in flight English runs on: RTX PRO 6000 | 82.6%79.2%–85.6% | 13.2%answered 97% | Not measureda plain answer line, no confidence | $1.764$1,764at the maker's list price | 74.21 per request32 in flight |
| 84 | Gemma 4 E4B default reasoning · chat Open · 8.0BLatency 8.39 sCost $0.199$199 per 1,000per 1M Gemma 4 E4B default reasoning · chat · Google · chat model Parameters: 8B dense · 4.5B effective, as the maker states it (counted from the released files: 8.00B). Accuracy in English 82.4% (95%: 79.0–85.4%), 547 kept reading questions Non-hallucination with “I don't know” 9.8% · answered 97.8% · error rate 17.2% Cost $0.199$199 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: not measured Latency: 8.39 s median per request, 32 in flight English runs on: RTX PRO 6000 | 82.4%79.0%–85.4% | 9.8%answered 98% | Not measureda plain answer line, no confidence | $0.199$199from GPU time | 8.391 per request32 in flight |
| 85 | Clef-Flash Open · 9.0BLatency 0.140 sCost $0.018$17.53 per 1,000per 1M Clef-Flash · Cloudflare · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.53B). Accuracy in English 82.3% (95%: 78.8–85.2%), 547 kept reading questions Non-hallucination with “I don't know” 5.3% · answered 99.1% · error rate 16.6% Cost $0.018$17.53 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.1141, Decision Score (AA formula) 85.43 Latency: 0.140 s median per request, 1 in flight English runs on: RTX PRO 6000 | 82.3%78.8%–85.2% | 5.3%answered 99% | 0.1141Decision Score 85.43 | $0.018$17.53at the maker's list price | 0.1401 per request1 in flight |
| 86 | Verdin-4B-v1.1 Open · 4.0BLatency 0.052 sCost $0.034$34.28 per 1,000per 1M Verdin-4B-v1.1 · victorbennett · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Accuracy in English 82.3% (95%: 78.8–85.2%), 547 kept reading questions Non-hallucination with “I don't know” 2.1% · answered 99.6% · error rate 17.2% Cost $0.034$34.28 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1032, Decision Score (AA formula) 85.97 Latency: 0.052 s median per request, 1 in flight English runs on: RTX PRO 6000 | 82.3%78.8%–85.2% | 2.1%answered 100% | 0.1032Decision Score 85.97 | $0.034$34.28from GPU time | 0.0521 per request1 in flight |
| 87 | Perplexity Decider Flash v1.2 Open · 4.0BLatency 0.037 sCost $0.004$3.93 per 1,000per 1M Perplexity Decider Flash v1.2 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Accuracy in English 81.9% (95%: 78.5–84.9%), 547 kept reading questions Non-hallucination with “I don't know” 4.8% · answered 99.1% · error rate 18.5% Cost $0.004$3.93 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: chosen-answer Brier 0.0982, Decision Score (AA formula) 86.04 Latency: 0.037 s median per request, 1 in flight English runs on: RTX PRO 6000 | 81.9%78.5%–84.9% | 4.8%answered 99% | 0.0982Decision Score 86.04 | $0.004$3.93at the maker's list price | 0.0371 per request1 in flight |
| 88 | KnowLine 4B Gen5.5 Open · 4.0BLatency 0.099 sCost No list price KnowLine 4B Gen5.5 · PelaAI · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy in English 80.8% (95%: 77.3–83.9%), 547 kept reading questions Non-hallucination with “I don't know” 0.9% · answered 99.8% · error rate 19.2% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1167, Decision Score (AA formula) 84.57 Latency: 0.099 s median per request, 1 in flight English runs on: DGX Spark | 80.8%77.3%–83.9% | 0.9%answered 100% | 0.1167Decision Score 84.57 | No list price | 0.0991 per request1 in flight |
| 89 | Qwen3.5-4B reasoning off · chat Open · 4.0BLatency 9.74 sCost $0.309$309 per 1,000per 1M Qwen3.5-4B reasoning off · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Accuracy in English 80.4% (95%: 76.9–83.5%), 547 kept reading questions Non-hallucination with “I don't know” 3.6% · answered 99.3% · error rate 19.5% Cost $0.309$309 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: not measured Latency: 9.74 s median per request, 32 in flight English runs on: RTX PRO 6000 | 80.4%76.9%–83.5% | 3.6%answered 99% | Not measureda plain answer line, no confidence | $0.309$309from GPU time | 9.741 per request32 in flight |
| 90 | Chinese Decision Pointer 4B Open · 4.0BLatency 0.038 sCost $0.027$27.21 per 1,000per 1M Chinese Decision Pointer 4B · quincyqiang · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Accuracy in English 80.4% (95%: 76.9–83.5%), 547 kept reading questions Non-hallucination with “I don't know” 3.7% · answered 99.3% · error rate 19.2% Cost $0.027$27.21 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1111, Decision Score (AA formula) 84.66 Latency: 0.038 s median per request, 1 in flight English runs on: RTX PRO 6000 | 80.4%76.9%–83.5% | 3.7%answered 99% | 0.1111Decision Score 84.66 | $0.027$27.21from GPU time | 0.0381 per request1 in flight |
| 91 | Qwen3-14B reasoning off · chat Open · 15BLatency 2.86 sCost $0.031$31.17 per 1,000per 1M Qwen3-14B reasoning off · chat · Alibaba · chat model Parameters: 14.8B dense, as the maker states it (counted from the released files: 14.77B). Accuracy in English 79.7% (95%: 76.1–82.9%), 547 kept reading questions Non-hallucination with “I don't know” 1.8% · answered 99.6% · error rate 19.8% Cost $0.031$31.17 per 1,000 questionsper 1M questions (at a host's list price for the same weights) · 16 questions per request Latency: 2.86 s median per request, 1 in flight English runs on: RTX PRO 6000 | 79.7%76.1%–82.9% | 1.8%answered 100% | Not measured | $0.031$31.17at a host's list price for the same weights | 2.8616 per request1 in flight |
| 92 | MiniCPM5-2B reasoning on · chat Open · 2.5BLatency 17.5 sCost $0.714$714 per 1,000per 1M MiniCPM5-2B reasoning on · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Accuracy in English 78.1%, the mean of 3 seeds (76.6%–79.0% min–max of 3 seeds), 547 kept reading questions Seeds: seed 0: run 1051, 419 of 547 right; seed 1: run 1118, 431 of 547 right; seed 2: run 1119, 432 of 547 right. Baseline server capacity not retained; repeats use max-num-seqs 16. Non-hallucination with “I don't know” 19.7% · answered 94.1% · error rate 22.1% (from run 1052) Cost $0.714$714 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request (from run 1051) English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line ANSWER: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking on; output cap 32,768 Confidence: not measured Latency: 17.5 s median per request, 32 in flight (from run 1051) English runs on: RTX PRO 6000 | 78.1%76.6%–79.0% min–max of 3 seeds | 19.7%answered 94%from run 1052 | Not measureda plain answer line, no confidence | $0.714$714from GPU time | 17.51 per request32 in flight |
| 93 | Metask-Jev-4B Open · 4.5BLatency 0.085 sCost $0.059$59.11 per 1,000per 1M Metask-Jev-4B · Metask Lab · decision model Parameters: 4.54B dense, as the maker states it (counted from the released files: 4.54B). Accuracy in English 77.1% (95%: 73.4–80.5%), 547 kept reading questions Non-hallucination with “I don't know” 1.6% · answered 99.6% · error rate 22.6% Cost $0.059$59.11 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1273, Decision Score (AA formula) 82.21 Latency: 0.085 s median per request, 1 in flight English runs on: RTX PRO 6000 | 77.1%73.4%–80.5% | 1.6%answered 100% | 0.1273Decision Score 82.21 | $0.059$59.11from GPU time | 0.0851 per request1 in flight |
| 94 | Gemma 4 E2B reasoning on · chat Open · 5.1BLatency 7.61 sCost $0.187$187 per 1,000per 1M Gemma 4 E2B reasoning on · chat · Google · chat model Parameters: 5.1B dense · 2.3B effective, as the maker states it (counted from the released files: 5.12B). Accuracy in English 76.2% (95%: 72.5–79.6%), 547 kept reading questions Non-hallucination with “I don't know” 11.3% · answered 96.3% · error rate 25.2% Cost $0.187$187 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E2B-it@4adf157), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0, thinking on; output cap 32,768 Confidence: not measured Latency: 7.61 s median per request, 32 in flight English runs on: RTX PRO 6000 | 76.2%72.5%–79.6% | 11.3%answered 96% | Not measureda plain answer line, no confidence | $0.187$187from GPU time | 7.611 per request32 in flight |
| 95 | Qwen3.5-2B reasoning on · chat Open · 2.0BLatency 54.4 sCost $1.361$1,361 per 1,000per 1M Qwen3.5-2B reasoning on · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy in English 75.9% (95%: 72.1–79.3%), 547 kept reading questions Non-hallucination with “I don't know” 9.9% · answered 97.1% · error rate 24.1% Cost $1.361$1,361 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line ANSWER: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768 Confidence: not measured Latency: 54.4 s median per request, 32 in flight English runs on: RTX PRO 6000 | 75.9%72.1%–79.3% | 9.9%answered 97% | Not measureda plain answer line, no confidence | $1.361$1,361from GPU time | 54.41 per request32 in flight |
| 96 | Gemma 4 E2B default reasoning · chat Open · 5.1BLatency 4.84 sCost $0.115$115 per 1,000per 1M Gemma 4 E2B default reasoning · chat · Google · chat model Parameters: 5.1B dense · 2.3B effective, as the maker states it (counted from the released files: 5.12B). Accuracy in English 74.6% (95%: 70.8–78.1%), 547 kept reading questions Non-hallucination with “I don't know” 11.5% · answered 96.5% · error rate 24.8% Cost $0.115$115 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E2B-it@4adf157), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0; output cap 8,192 Confidence: not measured Latency: 4.84 s median per request, 32 in flight English runs on: RTX PRO 6000 | 74.6%70.8%–78.1% | 11.5%answered 97% | Not measureda plain answer line, no confidence | $0.115$115from GPU time | 4.841 per request32 in flight |
| 97 | Granite-4.2-3B reasoning on · chat Open · 3.0BLatency 40.5 sCost $2.717$2,717 per 1,000per 1M Granite-4.2-3B reasoning on · chat · IBM · chat model Base model: Granite-4.1-3B-Base Parameters: 3B dense, as the maker states it (counted from the released files: 3.66B). Accuracy in English 74.4% (95%: 70.6–77.9%), 547 kept reading questions Non-hallucination with “I don't know” 16.1% · answered 95.8% · error rate 22.9% Cost $2.717$2,717 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, vLLM v0.30.0 BF16, revision ibm-granite/granite-4.2-3b@4e9c275), one question per request, plain answer line ANSWER: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking on; output cap 65,536 Confidence: not measured Latency: 40.5 s median per request, 32 in flight English runs on: RTX PRO 6000 | 74.4%70.6%–77.9% | 16.1%answered 96% | Not measureda plain answer line, no confidence | $2.717$2,717from GPU time | 40.51 per request32 in flight |
| 98 | LFM2.5-2.6B default reasoning · chat Open · 2.7BLatency 15.0 sCost $0.964$964 per 1,000per 1M LFM2.5-2.6B default reasoning · chat · Liquid AI · chat model Parameters: 2.69B dense, as the maker states it (counted from the released files: 2.70B). Accuracy in English 73.9% (95%: 70.0–77.4%), 547 kept reading questions Non-hallucination with “I don't know” 15.7% · answered 95.6% · error rate 21.6% Cost $0.964$964 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Confidence: not measured Latency: 15.0 s median per request, 32 in flight English runs on: RTX PRO 6000 | 73.9%70.0%–77.4% | 15.7%answered 96% | Not measureda plain answer line, no confidence | $0.964$964at the maker's list price | 15.01 per request32 in flight |
| 99 | Erk-14B reasoning off · chat Open · 14BLatency 2.86 sCost $0.123$123 per 1,000per 1M Erk-14B reasoning off · chat · eCloud Technology · chat model Parameters: 14B dense, as the maker states it (counted from the released files: 14.77B). Accuracy in English 72.0% (95%: 68.1–75.6%), 547 kept reading questions Non-hallucination with “I don't know” 2.1% · answered 99.5% · error rate 25.9% Cost $0.123$123 per 1,000 questionsper 1M questions (from GPU time) · 16 questions per request Latency: 2.86 s median per request, 1 in flight English runs on: RTX PRO 6000 | 72.0%68.1%–75.6% | 2.1%answered 99% | Not measured | $0.123$123from GPU time | 2.8616 per request1 in flight |
| 100 | LFM3-1.7B reasoning off · chat Open · 1.7BLatency 4.69 sCost $0.355$355 per 1,000per 1M LFM3-1.7B reasoning off · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Accuracy in English 71.8%, the mean of 3 seeds (70.9%–72.9% min–max of 3 seeds), 547 kept reading questions Seeds: seed 0: run 1223, 399 of 547 right; seed 1: run 1302, 388 of 547 right; seed 2: run 1303, 391 of 547 right Non-hallucination with “I don't know” 11.6% · answered 96.7% · error rate 25.9% (from run 1225) Cost $0.355$355 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request (from run 1223) Confidence: not measured Latency: 4.69 s median per request, 32 in flight (from run 1223) English runs on: RTX PRO 6000 | 71.8%70.9%–72.9% min–max of 3 seeds | 11.6%answered 97%from run 1225 | Not measureda plain answer line, no confidence | $0.355$355at the maker's list price | 4.691 per request32 in flight |
| 101 | KnowLine-2B-Gen1 Open · 2.0BLatency 0.062 sCost No list price KnowLine-2B-Gen1 · PelaAI · decision model Base model: Qwen3.5-2B Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy in English 71.3% (95%: 67.4–74.9%), 547 kept reading questions Non-hallucination with “I don't know” 22.0% · answered 93.1% · error rate 26.5% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1567, Decision Score (AA formula) 77.82 Latency: 0.062 s median per request, 1 in flight English runs on: DGX Spark | 71.3%67.4%–74.9% | 22.0%answered 93% | 0.1567Decision Score 77.82 | No list price | 0.0621 per request1 in flight |
| 102 | LFM3-1.7B reasoning on · chat Open · 1.7BLatency 5.24 sCost $0.387$387 per 1,000per 1M LFM3-1.7B reasoning on · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Accuracy in English 71.1%, the mean of 3 seeds (70.4%–71.7% min–max of 3 seeds), 547 kept reading questions Seeds: seed 0: run 1091, 392 of 547 right; seed 1: run 1115, 390 of 547 right; seed 2: run 1116, 385 of 547 right Non-hallucination with “I don't know” 15.6% · answered 95.2% · error rate 27.1% (from run 1062) Cost $0.387$387 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request (from run 1061) Confidence: not measured Latency: 5.24 s median per request, 32 in flight (from run 1061) English runs on: RTX PRO 6000 | 71.1%70.4%–71.7% min–max of 3 seeds | 15.6%answered 95%from run 1062 | Not measureda plain answer line, no confidence | $0.387$387at the maker's list price | 5.241 per request32 in flight |
| 103 | LFM2.5-8B-A1B default reasoning · chat Open · 8.3B MoE · 1.5B activeLatency 19.4 sCost $0.517$517 per 1,000per 1M LFM2.5-8B-A1B default reasoning · chat · Liquid AI · chat model Mixture of experts: 8.3B total (counted from the released files: 8.47B), about 1.5B active per step. Size class uses the total (8.3B). Accuracy in English 70.4% (95%: 66.4–74.1%), 547 kept reading questions Non-hallucination with “I don't know” 26.4% · answered 90.5% · error rate 28.7% Cost $0.517$517 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: not measured Latency: 19.4 s median per request, 32 in flight English runs on: RTX PRO 6000 | 70.4%66.4%–74.1% | 26.4%answered 90% | Not measureda plain answer line, no confidence | $0.517$517from GPU time | 19.41 per request32 in flight |
| 104 | GLiDE 2B Open · 1.9BLatency 0.035 sCost No list price GLiDE 2B · Fastino · decision model Parameters: 1.88B dense, as the maker states it (counted from the released files: 1.88B). Accuracy in English 70.0% (95%: 66.0–73.7%), 547 kept reading questions Non-hallucination with “I don't know” 26.3% · answered 91.4% · error rate 26.4% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1626, Decision Score (AA formula) 76.88 Latency: 0.035 s median per request, 1 in flight English runs on: DGX Spark | 70.0%66.0%–73.7% | 26.3%answered 91% | 0.1626Decision Score 76.88 | No list price | 0.0351 per request1 in flight |
| 105 | StartLux-Decision-v1.1-2B-Instant Open · 2.0BLatency 0.040 sCost No list price StartLux-Decision-v1.1-2B-Instant · StartLux · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy in English 69.7% (95%: 65.7–73.4%), 547 kept reading questions Non-hallucination with “I don't know” 16.9% · answered 94.1% · error rate 30.5% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1633, Decision Score (AA formula) 76.66 Latency: 0.040 s median per request, 1 in flight English runs on: DGX Spark | 69.7%65.7%–73.4% | 16.9%answered 94% | 0.1633Decision Score 76.66 | No list price | 0.0401 per request1 in flight |
| 106 | MiniCPM5-2B reasoning off · chat Open · 2.5BLatency 7.29 sCost $0.429$429 per 1,000per 1M MiniCPM5-2B reasoning off · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Accuracy in English 67.8%, the mean of 3 seeds (66.7%–68.9% min–max of 3 seeds), 547 kept reading questions Seeds: seed 0: run 1094, 370 of 547 right; seed 1: run 1300, 377 of 547 right; seed 2: run 1301, 365 of 547 right. Seeds 1 and 2 on logic's load of 15 November (vLLM v0.30.0, max-num-seqs 16), which logic gives as run 1094's setup (#9335); the server's settings are not recorded with a run. Non-hallucination with “I don't know” 16.8% · answered 89.2% · error rate 33.6% (from run 1095) Cost $0.429$429 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request (from run 1094) Confidence: not measured Latency: 7.29 s median per request, 32 in flight (from run 1094) English runs on: RTX PRO 6000 | 67.8%66.7%–68.9% min–max of 3 seeds | 16.8%answered 89%from run 1095 | Not measureda plain answer line, no confidence | $0.429$429from GPU time | 7.291 per request32 in flight |
| 107 | Qwen3.5-2B reasoning off · chat Open · 2.0BLatency 9.27 sCost $0.280$280 per 1,000per 1M Qwen3.5-2B reasoning off · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Accuracy in English 67.3% (95%: 63.2–71.1%), 547 kept reading questions Non-hallucination with “I don't know” 10.3% · answered 96.3% · error rate 29.6% Cost $0.280$280 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: not measured Latency: 9.27 s median per request, 32 in flight English runs on: RTX PRO 6000 | 67.3%63.2%–71.1% | 10.3%answered 96% | Not measureda plain answer line, no confidence | $0.280$280from GPU time | 9.271 per request32 in flight |
| 108 | d1-2B Open · 1.9BLatency 0.011 sCost $0.008$8.08 per 1,000per 1M d1-2B · Liquid AI · decision model Base model: LFM3-VL-2B Parameters: 1.92B dense, as the maker states it (counted from the released files: 1.92B). Accuracy in English 65.6% (95%: 61.6–69.5%), 547 kept reading questions Non-hallucination with “I don't know” 60.9% · answered 72.4% · error rate 24.5% Cost $0.008$8.08 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1731, Decision Score (AA formula) 74.16 Latency: 0.011 s median per request, 1 in flight English runs on: RTX PRO 6000 | 65.6%61.6%–69.5% | 60.9%answered 72% | 0.1731Decision Score 74.16 | $0.008$8.08from GPU time | 0.0111 per request1 in flight |
| 109 | GLiDE 0.8B Open · 0.8BLatency 0.023 sCost No list price GLiDE 0.8B · Fastino · decision model Parameters: 750M dense, as the maker states it (counted from the released files: 0.75B). Accuracy in English 60.7% (95%: 56.5–64.7%), 547 kept reading questions Non-hallucination with “I don't know” 21.7% · answered 91.0% · error rate 35.5% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1951, Decision Score (AA formula) 70.59 Latency: 0.023 s median per request, 1 in flight English runs on: DGX Spark | 60.7%56.5%–64.7% | 21.7%answered 91% | 0.1951Decision Score 70.59 | No list price | 0.0231 per request1 in flight |
| 110 | StartLux-Decision-v1.1-0.8B-Instant Open · 0.8BLatency 0.028 sCost No list price StartLux-Decision-v1.1-0.8B-Instant · StartLux · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Accuracy in English 59.6% (95%: 55.4–63.6%), 547 kept reading questions Non-hallucination with “I don't know” 5.2% · answered 97.8% · error rate 40.6% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1942, Decision Score (AA formula) 70.09 Latency: 0.028 s median per request, 1 in flight English runs on: DGX Spark | 59.6%55.4%–63.6% | 5.2%answered 98% | 0.1942Decision Score 70.09 | No list price | 0.0281 per request1 in flight |
| 111 | GLiDE 0.4B Open · 0.4BLatency 0.014 sCost No list price GLiDE 0.4B · Fastino · decision model Parameters: 400M dense, as the maker states it (counted from the released files: 0.40B). Accuracy in English 59.4% (95%: 55.2–63.5%), 547 kept reading questions Non-hallucination with “I don't know” 53.1% · answered 72.0% · error rate 34.3% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1973, Decision Score (AA formula) 69.84 Latency: 0.014 s median per request, 1 in flight English runs on: DGX Spark | 59.4%55.2%–63.5% | 53.1%answered 72% | 0.1973Decision Score 69.84 | No list price | 0.0141 per request1 in flight |
| 112 | OneJev1.6-0.8B Open · 0.8BLatency 0.018 sCost No list price OneJev1.6-0.8B · OmniJev · decision model Parameters: 800M dense, as the maker states it (counted from the released files: 0.85B). Accuracy in English 59.4% (95%: 55.2–63.5%), 547 kept reading questions Non-hallucination with “I don't know” 28.4% · answered 86.5% · error rate 39.5% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1978, Decision Score (AA formula) 69.82 Latency: 0.018 s median per request, 1 in flight English runs on: DGX Spark | 59.4%55.2%–63.5% | 28.4%answered 86% | 0.1978Decision Score 69.82 | No list price | 0.0181 per request1 in flight |
| 113 | KnowLine-0.8B-Gen1 Open · 0.8BLatency 0.033 sCost No list price KnowLine-0.8B-Gen1 · PelaAI · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Accuracy in English 58.9% (95%: 54.7–62.9%), 547 kept reading questions Non-hallucination with “I don't know” 10.7% · answered 95.2% · error rate 41.7% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.1956, Decision Score (AA formula) 69.65 Latency: 0.033 s median per request, 1 in flight English runs on: DGX Spark | 58.9%54.7%–62.9% | 10.7%answered 95% | 0.1956Decision Score 69.65 | No list price | 0.0331 per request1 in flight |
| 114 | GLiClass v4 Edge Open · 0.4BLatency 0.017 sCost No list price GLiClass v4 Edge · Knowledgator · decision model Base model: DeBERTa-v3-large Parameters: 400M dense, as the maker states it (counted from the released files: 0.44B). Accuracy in English 58.3% (95%: 54.1–62.4%), 547 kept reading questions Non-hallucination with “I don't know” 21.0% · answered 90.5% · error rate 39.6% Cost: No list price · 1 question per request Confidence: chosen-answer Brier 0.2021, Decision Score (AA formula) 69.05 Latency: 0.017 s median per request, 1 in flight English runs on: DGX Spark | 58.3%54.1%–62.4% | 21.0%answered 90% | 0.2021Decision Score 69.05 | No list price | 0.0171 per request1 in flight |
| 115 | Decider 1 Closed · EfficientLatency 0.699 sCost $0.015$14.95 per 1,000per 1M Decider 1 · meraGPT · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://meragpt.com/docs/models, read 2026-10-19) Accuracy in English 55.8% (95%: 51.6–59.9%), 547 kept reading questions Cost $0.015$14.95 per 1,000 questionsper 1M questions (at list prices) · 1 question per request Confidence: chosen-answer Brier 0.2082, Decision Score (AA formula) 67.47 Latency: 0.699 s median per request, 1 in flight English runs on: hosted API | 55.8%51.6%–59.9% | Not run | 0.2082Decision Score 67.47 | $0.015$14.95at list prices | 0.6991 per request1 in flight |
| 116 | Raon-1 nano reasoning off · chat Open · 3.1BLatency 3.74 sCost $0.235$235 per 1,000per 1M Raon-1 nano reasoning off · chat · KRAFTON · chat model Base model: Raon-1 mini preview (MoE-to-Dense pruning and distillation) Parameters: 3.1B dense, as the maker states it (counted from the released files: 3.53B). Accuracy in English 53.7% (95%: 49.6–57.9%), 547 kept reading questions Non-hallucination with “I don't know” 26.2% · answered 86.8% · error rate 42.1% Cost $0.235$235 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request English only: no Text tab run, so not in the Turkish and Both parts How we called it: our GPU (logic's server, KRAFTON's vLLM fork (krafton-ai/vllm@13a5dd8; not a vLLM release) BF16, revision KRAFTON/Raon-1-nano@fcf2d1d), one question per request, plain answer line ANSWER: X (no answer schema); sampling presence_penalty 1.5, seed 0, temperature 0.1, thinking off; output cap 32,768 Confidence: not measured Latency: 3.74 s median per request, 32 in flight English runs on: RTX PRO 6000 | 53.7%49.6%–57.9% | 26.2%answered 87% | Not measureda plain answer line, no confidence | $0.235$235from GPU time | 3.741 per request32 in flight |
| 117 | Strands Decider 2B Open · 2.0BLatency 0.026 sCost $0.018$18.15 per 1,000per 1M Strands Decider 2B · Strands Agents · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.29B). Accuracy in English 45.5% (95%: 41.4–49.7%), 547 kept reading questions Non-hallucination with “I don't know” 0.7% · answered 99.6% · error rate 55.2% Cost $0.018$18.15 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1736, Decision Score (AA formula) 64.08 Latency: 0.026 s median per request, 1 in flight English runs on: RTX PRO 6000 | 45.5%41.4%–49.7% | 0.7%answered 100% | 0.1736Decision Score 64.08 | $0.018$18.15from GPU time | 0.0261 per request1 in flight |
| 118 | Laya Open · 0.4BLatency 0.011 sCost $0.008$7.84 per 1,000per 1M Laya · Convai Innovations · decision model Parameters: 421M dense, as the maker states it (counted from the released files: 0.42B). Accuracy in English 22.9% (95%: 19.5–26.6%), 547 kept reading questions Non-hallucination with “I don't know” 18.4% · answered 84.8% · error rate 79.1% Cost $0.008$7.84 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.1971, Decision Score (AA formula) 51.57 Latency: 0.011 s median per request, 1 in flight English runs on: RTX PRO 6000 | 22.9%19.5%–26.6% | 18.4%answered 85% | 0.1971Decision Score 51.57 | $0.008$7.84from GPU time | 0.0111 per request1 in flight |
| 119 | GLiNER2.5-multi-Decide Open · 0.3BLatency 0.009 sCost $0.007$6.73 per 1,000per 1M GLiNER2.5-multi-Decide · Fastino · decision model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Accuracy in English 21.2% (95%: 18.0–24.8%), 547 kept reading questions Non-hallucination with “I don't know” 16.2% · answered 87.2% · error rate 76.1% Cost $0.007$6.73 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.2604, Decision Score (AA formula) 47.58 Latency: 0.009 s median per request, 1 in flight English runs on: RTX PRO 6000 | 21.2%18.0%–24.8% | 16.2%answered 87% | 0.2604Decision Score 47.58 | $0.007$6.73from GPU time | 0.0091 per request1 in flight |
| 120 | GLiNER2.5 Multi Open · 0.3BLatency 0.009 sCost $0.008$8.39 per 1,000per 1M GLiNER2.5 Multi · Fastino · chat model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Accuracy in English 16.3% (95%: 13.4–19.6%), 547 kept reading questions Non-hallucination with “I don't know” 19.6% · answered 83.2% · error rate 83.1% Cost $0.008$8.39 per 1,000 questionsper 1M questions (at the maker's list price) · 1 question per request Latency: 0.009 s median per request, 1 in flight English runs on: RTX PRO 6000 | 16.3%13.4%–19.6% | 19.6%answered 83% | Not measured | $0.008$8.39at the maker's list price | 0.0091 per request1 in flight |
| 121 | Laya Multilingual Open · 0.3BLatency 0.018 sCost $0.013$12.62 per 1,000per 1M Laya Multilingual · Convai Innovations · decision model Parameters: 322M dense, as the maker states it (counted from the released files: 0.32B). Accuracy in English 15.4% (95%: 12.6–18.6%), 547 kept reading questions Non-hallucination with “I don't know” 42.6% · answered 62.0% · error rate 82.6% Cost $0.013$12.62 per 1,000 questionsper 1M questions (from GPU time) · 1 question per request Confidence: chosen-answer Brier 0.3671, Decision Score (AA formula) 39.32 Latency: 0.018 s median per request, 1 in flight English runs on: RTX PRO 6000 | 15.4%12.6%–18.6% | 42.6%answered 62% | 0.3671Decision Score 39.32 | $0.013$12.62from GPU time | 0.0181 per request1 in flight |
Accuracy in English: the share of the 547 kept reading questions, translated, answered right, with its 95% interval as the whisker; a setup run with three seeds shows their mean, without a whisker, its lowest and highest under it; higher is better.
No row of this part matches these keys. The small open chat models ran in English only.
Rank: the row's place among the rows in view, by accuracy, 1, 2, 3 without repeats; rows with the same score keep the table's order, which claims no difference between them. No all-pairs test covers the English reading part, so it has no tiers. A setup with web search is not ranked (Rank: –).
| Rank | Model | Turkish · English547 questions, 95% intervals | Turkishof 547 kept | Englishof 547 kept | Differencepoints, English minus Turkish |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 max · chat Closed · Flagship Claude Fable 5.2 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $8.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-11-11) Turkish 97.8% (95%: 96.2–98.7%) · English 98.0% (95%: 96.4–98.9%), on the same 547 kept reading questions English minus Turkish: +0.2 points | 97.8%98.0% | 97.8% | 98.0% | +0.2 |
| 2 | Gemini 4 Carbon high · chat Closed · Flagship Gemini 4 Carbon high · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish 97.6% (95%: 96.0–98.6%) · English 94.3% (95%: 92.1–96.0%), on the same 547 kept reading questions English minus Turkish: -3.3 points | 97.6%94.3% | 97.6% | 94.3% | -3.3 |
| 3 | Gemini 4 Hydrogen high · chat Closed · Workhorse Gemini 4 Hydrogen high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-11-12) Turkish 97.1% (95%: 95.3–98.2%) · English 96.7% (95%: 94.9–97.9%), on the same 547 kept reading questions English minus Turkish: -0.4 points | 97.1%96.7% | 97.1% | 96.7% | -0.4 |
| 4 | Gemini 3.8 Flash high · chat Closed · Workhorse Gemini 3.8 Flash high · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish 96.9% (95%: 95.1–98.1%) · English 96.0% (95%: 94.0–97.3%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 96.9%96.0% | 96.9% | 96.0% | -0.9 |
| Gemini 3.8 Flash low · chat Closed · Workhorseother setting of Gemini 3.8 Flash Gemini 3.8 Flash low · chat · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish 95.2% (95%: 93.1–96.7%) · English 94.3% (95%: 92.1–96.0%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 95.2%94.3% | 95.2% | 94.3% | -0.9 | |
| Gemini 3.8 Flash low · chat · 1 per request Closed · Workhorseother setting of Gemini 3.8 Flash Gemini 3.8 Flash low · chat · 1 per request · Google · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-08) Turkish 96.2% (95%: 94.2–97.5%) · English 95.2% (95%: 93.1–96.7%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 96.2%95.2% | 96.2% | 95.2% | -0.9 | |
| 5 | GPT-6 Astra high · chat Closed · Flagship GPT-6 Astra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 96.9% (95%: 95.1–98.1%) · English 96.2% (95%: 94.2–97.5%), on the same 547 kept reading questions English minus Turkish: -0.7 points | 96.9%96.2% | 96.9% | 96.2% | -0.7 |
| GPT-6 Astra medium · chat Closed · Flagshipother setting of GPT-6 Astra GPT-6 Astra medium · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 96.7% (95%: 94.9–97.9%) · English 96.0% (95%: 94.0–97.3%), on the same 547 kept reading questions English minus Turkish: -0.7 points | 96.7%96.0% | 96.7% | 96.0% | -0.7 | |
| GPT-6 Astra low · chat Closed · Flagshipother setting of GPT-6 Astra GPT-6 Astra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 96.7% (95%: 94.9–97.9%) · English 96.3% (95%: 94.4–97.6%), on the same 547 kept reading questions English minus Turkish: -0.4 points | 96.7%96.3% | 96.7% | 96.3% | -0.4 | |
| GPT-6 Astra low · chat · 1 per request Closed · Flagshipother setting of GPT-6 Astra GPT-6 Astra low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 97.1% (95%: 95.3–98.2%) · English 95.6% (95%: 93.6–97.0%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 97.1%95.6% | 97.1% | 95.6% | -1.5 | |
| 6 | Gemini 4 Argon low · chat Closed · Flagship Gemini 4 Argon low · chat · Google · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://ai.google.dev/gemini-api/docs/pricing, read 2026-10-28) Turkish 96.9% (95%: 95.1–98.1%) · English 96.0% (95%: 94.0–97.3%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 96.9%96.0% | 96.9% | 96.0% | -0.9 |
| 7 | GPT-6.1 Terra high · chat Closed · Workhorse GPT-6.1 Terra high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish 96.3% (95%: 94.4–97.6%) · English 96.2% (95%: 94.2–97.5%), on the same 547 kept reading questions English minus Turkish: -0.2 points | 96.3%96.2% | 96.3% | 96.2% | -0.2 |
| GPT-6.1 Terra low · chat Closed · Workhorseother setting of GPT-6.1 Terra GPT-6.1 Terra low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-28) Turkish 95.6% (95%: 93.6–97.0%) · English 96.2% (95%: 94.2–97.5%), on the same 547 kept reading questions English minus Turkish: +0.5 points | 95.6%96.2% | 95.6% | 96.2% | +0.5 | |
| 8 | GPT-6.1 Sol high · chat Closed · Workhorse GPT-6.1 Sol high · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 96.2% (95%: 94.2–97.5%) · English 95.6% (95%: 93.6–97.0%), on the same 547 kept reading questions English minus Turkish: -0.5 points | 96.2%95.6% | 96.2% | 95.6% | -0.5 |
| GPT-6.1 Sol low · chat Closed · Workhorseother setting of GPT-6.1 Sol GPT-6.1 Sol low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 95.8% (95%: 93.8–97.2%) · English 95.8% (95%: 93.8–97.2%), on the same 547 kept reading questions English minus Turkish: +0.0 points | 95.8%95.8% | 95.8% | 95.8% | +0.0 | |
| GPT-6.1 Sol low · chat · 1 per request Closed · Workhorseother setting of GPT-6.1 Sol GPT-6.1 Sol low · chat · 1 per request · OpenAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://developers.openai.com/api/docs/pricing, read 2026-10-08) Turkish 95.6% (95%: 93.6–97.0%) · English 95.2% (95%: 93.1–96.7%), on the same 547 kept reading questions English minus Turkish: -0.4 points | 95.6%95.2% | 95.6% | 95.2% | -0.4 | |
| 9 | Claude Opus 5.5 high · chat Closed · Flagship Claude Opus 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 95.6% (95%: 93.6–97.0%) · English 94.5% (95%: 92.3–96.1%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 95.6%94.5% | 95.6% | 94.5% | -1.1 |
| Claude Opus 5.5 low · chat Closed · Flagshipother setting of Claude Opus 5.5 Claude Opus 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 95.1% (95%: 92.9–96.6%) · English 93.1% (95%: 90.6–94.9%), on the same 547 kept reading questions English minus Turkish: -2.0 points | 95.1%93.1% | 95.1% | 93.1% | -2.0 | |
| Claude Opus 5.5 low · chat · 1 per request Closed · Flagshipother setting of Claude Opus 5.5 Claude Opus 5.5 low · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $4.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 95.4% (95%: 93.3–96.9%) · English 94.7% (95%: 92.5–96.3%), on the same 547 kept reading questions English minus Turkish: -0.7 points | 95.4%94.7% | 95.4% | 94.7% | -0.7 | |
| 10 | Claude Fable 5.1 high · chat Closed · Flagship Claude Fable 5.1 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Flagship, by its standard list price: $10.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-28) Turkish 95.4% (95%: 93.3–96.9%) · English 96.0% (95%: 94.0–97.3%), on the same 547 kept reading questions English minus Turkish: +0.5 points | 95.4%96.0% | 95.4% | 96.0% | +0.5 |
| 11 | Kimi K3.1 Pro low · chat Open · 2.8T MoE · 104B active Kimi K3.1 Pro low · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish 95.2% (95%: 93.1–96.7%) · English 92.3% (95%: 89.8–94.3%), on the same 547 kept reading questions English minus Turkish: -2.9 points | 95.2%92.3% | 95.2% | 92.3% | -2.9 |
| Kimi K3.1 Pro no reasoning · chat Open · 2.8T MoE · 104B activeother setting of Kimi K3.1 Pro Kimi K3.1 Pro no reasoning · chat · Moonshot · chat model Mixture of experts: 2.8T total, about 104B active per step. Size class uses the total (2.8T). Turkish 94.0% (95%: 91.6–95.7%) · English 91.4% (95%: 88.8–93.5%), on the same 547 kept reading questions English minus Turkish: -2.6 points | 94.0%91.4% | 94.0% | 91.4% | -2.6 | |
| 12 | Grok 4.7 high · chat Closed · Workhorse Grok 4.7 high · chat · xAI · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://docs.x.ai/docs/models/grok-4.7, read 2026-10-28) Turkish 95.1% (95%: 92.9–96.6%) · English 93.4% (95%: 91.0–95.2%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 95.1%93.4% | 95.1% | 93.4% | -1.6 |
| 13 | Karpo Decision 31B Open · 31B Karpo Decision 31B · MachinePulse · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 94.9% (95%: 92.7–96.4%) · English 92.1% (95%: 89.6–94.1%), on the same 547 kept reading questions English minus Turkish: -2.7 points | 94.9%92.1% | 94.9% | 92.1% | -2.7 |
| 14 | Qwen4-Max high · chat Closed · Workhorse Qwen4-Max high · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $1.50 per 1M input tokens (https://openrouter.ai/qwen/qwen4-max, read 2026-11-16) Turkish 94.7% (95%: 92.5–96.3%) · English 94.5% (95%: 92.3–96.1%), on the same 547 kept reading questions English minus Turkish: -0.2 points | 94.7%94.5% | 94.7% | 94.5% | -0.2 |
| 15 | DeepSeek-V4.1-Flash reasoning on · chat · 1017 Open · 552B MoE · 16B active DeepSeek-V4.1-Flash reasoning on · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish 94.5% (95%: 92.3–96.1%) · English 92.5% (95%: 90.0–94.4%), on the same 547 kept reading questions English minus Turkish: -2.0 points | 94.5%92.5% | 94.5% | 92.5% | -2.0 |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017 Open · 552B MoE · 16B activeother setting of DeepSeek-V4.1-Flash DeepSeek-V4.1-Flash reasoning off · chat · 1017 · DeepSeek · chat model Mixture of experts: 552B total, about 16B active per step. Size class uses the total (552B). Turkish 89.9% (95%: 87.1–92.2%) · English 87.0% (95%: 83.9–89.6%), on the same 547 kept reading questions English minus Turkish: -2.9 points | 89.9%87.0% | 89.9% | 87.0% | -2.9 | |
| 16 | Copernicus-31B Open · 31B Copernicus-31B · Poindexter Labs · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 94.5% (95%: 92.3–96.1%) · English 92.9% (95%: 90.4–94.7%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 94.5%92.9% | 94.5% | 92.9% | -1.6 |
| 17 | KnowLine-31B-Gen2 Open · 31B KnowLine-31B-Gen2 · PelaAI · decision model Base model: Gemma 4 31B Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 94.5% (95%: 92.3–96.1%) · English 92.3% (95%: 89.8–94.3%), on the same 547 kept reading questions English minus Turkish: -2.2 points | 94.5%92.3% | 94.5% | 92.3% | -2.2 |
| 18 | KnowLine-31B-Gen1.5 Open · 31B KnowLine-31B-Gen1.5 · PelaAI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 94.3% (95%: 92.1–96.0%) · English 93.6% (95%: 91.2–95.4%), on the same 547 kept reading questions English minus Turkish: -0.7 points | 94.3%93.6% | 94.3% | 93.6% | -0.7 |
| 19 | REAX 1.3 Open · 57B REAX 1.3 · REAX · decision model Two models run together, option probabilities averaged 50/50: 57.36B in all, all used on every question. Turkish 94.0% (95%: 91.6–95.7%) · English 92.5% (95%: 90.0–94.4%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 94.0%92.5% | 94.0% | 92.5% | -1.5 |
| 20 | d1-ultra Closed · Efficient d1-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-ultra, read 2026-10-16) Turkish 93.6% (95%: 91.2–95.4%) · English 90.7% (95%: 87.9–92.8%), on the same 547 kept reading questions English minus Turkish: -2.9 points | 93.6%90.7% | 93.6% | 90.7% | -2.9 |
| 21 | Turion Jev 31B Open · 31B Turion Jev 31B · Turion.AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 93.6% (95%: 91.2–95.4%) · English 90.7% (95%: 87.9–92.8%), on the same 547 kept reading questions English minus Turkish: -2.9 points | 93.6%90.7% | 93.6% | 90.7% | -2.9 |
| 22 | Logits Sigma-1 Duo Open · 57B Logits Sigma-1 Duo · Logits · decision model Parameters: 57.36B dense, as the maker states it. Turkish 93.4% (95%: 91.0–95.2%) · English 91.0% (95%: 88.4–93.2%), on the same 547 kept reading questions English minus Turkish: -2.4 points | 93.4%91.0% | 93.4% | 91.0% | -2.4 |
| 23 | Sieve 31B v3 Open · 31B Sieve 31B v3 · Sthānika AI · decision model Parameters: 31B dense, as the maker states it (counted from the released files: 31.27B). Turkish 93.4% (95%: 91.0–95.2%) · English 90.7% (95%: 87.9–92.8%), on the same 547 kept reading questions English minus Turkish: -2.7 points | 93.4%90.7% | 93.4% | 90.7% | -2.7 |
| 24 | Mammoth-1 Open · 27B Mammoth-1 · Base Labs · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish 92.9% (95%: 90.4–94.7%) · English 90.1% (95%: 87.3–92.4%), on the same 547 kept reading questions English minus Turkish: -2.7 points | 92.9%90.1% | 92.9% | 90.1% | -2.7 |
| 25 | Gemini Decision 1 20261111 Closed · Efficient Gemini Decision 1 20261111 · Google · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/google/gemini-decision-1, read 2026-11-11) Turkish 92.9% (95%: 90.4–94.7%) · English 91.0% (95%: 88.4–93.2%), on the same 547 kept reading questions English minus Turkish: -1.8 points | 92.9%91.0% | 92.9% | 91.0% | -1.8 |
| 26 | Gemma 4 31B reasoning off · chat Open · 31B Gemma 4 31B reasoning off · chat · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish 92.5% (95%: 90.0–94.4%) · English 90.5% (95%: 87.7–92.7%), on the same 547 kept reading questions English minus Turkish: -2.0 points | 92.5%90.5% | 92.5% | 90.5% | -2.0 |
| Gemma 4 31B reasoning off · chat · 1 per request Open · 31Bother setting of Gemma 4 31B Gemma 4 31B reasoning off · chat · 1 per request · Google · chat model Parameters: 30.7B dense, as the maker states it (counted from the released files: 31.27B). Turkish 90.3% (95%: 87.5–92.5%) · English 89.9% (95%: 87.1–92.2%), on the same 547 kept reading questions English minus Turkish: -0.4 points | 90.3%89.9% | 90.3% | 89.9% | -0.4 | |
| 27 | Perplexity Decider 27B v1.2 DGX Spark Open · 27B Perplexity Decider 27B v1.2 DGX Spark · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.4% (95%: 88.8–93.5%) · English 88.7% (95%: 85.7–91.1%), on the same 547 kept reading questions English minus Turkish: -2.7 points | 91.4%88.7% | 91.4% | 88.7% | -2.7 |
| Perplexity Decider 27B v1.2 API Open · 27Bother setting of Perplexity Decider 27B v1.2 Perplexity Decider 27B v1.2 API · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.0% (95%: 88.4–93.2%) · English 88.5% (95%: 85.5–90.9%), on the same 547 kept reading questions English minus Turkish: -2.6 points | 91.0%88.5% | 91.0% | 88.5% | -2.6 | |
| 28 | d3-ultra Open · 26B d3-ultra · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.4% (95%: 88.8–93.5%) · English 89.4% (95%: 86.5–91.7%), on the same 547 kept reading questions English minus Turkish: -2.0 points | 91.4%89.4% | 91.4% | 89.4% | -2.0 |
| 29 | Drex Max Open · 27B Drex Max · Nace.AI · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.4% (95%: 88.8–93.5%) · English 89.6% (95%: 86.7–91.9%), on the same 547 kept reading questions English minus Turkish: -1.8 points | 91.4%89.6% | 91.4% | 89.6% | -1.8 |
| 30 | Qwen3.5-397B-A17B high · chat Open · 397B MoE · 17B active Qwen3.5-397B-A17B high · chat · Alibaba · chat model Mixture of experts: 397B total, about 17B active per step. Size class uses the total (397B). Turkish 91.4% (95%: 88.8–93.5%) · English 91.4% (95%: 88.8–93.5%), on the same 547 kept reading questions English minus Turkish: +0.0 points | 91.4%91.4% | 91.4% | 91.4% | +0.0 |
| 31 | GLiDE 27B Open · 27B GLiDE 27B · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.2% (95%: 88.6–93.3%) · English 90.9% (95%: 88.2–93.0%), on the same 547 kept reading questions English minus Turkish: -0.4 points | 91.2%90.9% | 91.2% | 90.9% | -0.4 |
| 32 | d3 Open · 26B d3 · vLLM Semantic Router · decision model Parameters: 26.09B dense, as the maker states it (counted from the released files: 26.09B). Turkish 91.0% (95%: 88.4–93.2%) · English 89.8% (95%: 86.9–92.0%), on the same 547 kept reading questions English minus Turkish: -1.3 points | 91.0%89.8% | 91.0% | 89.8% | -1.3 |
| 33 | Muse Ember max · chat Closed · Efficient Muse Ember max · chat · Meta · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/meta/muse-ember, read 2026-11-10) Turkish 90.9% (95%: 88.2–93.0%) · English 89.6% (95%: 86.7–91.9%), on the same 547 kept reading questions English minus Turkish: -1.3 points | 90.9%89.6% | 90.9% | 89.6% | -1.3 |
| 34 | d1.2-ultra Closed · Efficient d1.2-ultra · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish 90.9% (95%: 88.2–93.0%) · English 90.3% (95%: 87.5–92.5%), on the same 547 kept reading questions English minus Turkish: -0.5 points | 90.9%90.3% | 90.9% | 90.3% | -0.5 |
| 35 | Qwen3.8-27B low · chat Open · 27B Qwen3.8-27B low · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish 90.7% (95%: 87.9–92.8%) · English 89.0% (95%: 86.1–91.4%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 90.7%89.0% | 90.7% | 89.0% | -1.6 |
| Qwen3.8-27B yes/no scoring Open · 27Bother setting of Qwen3.8-27B Qwen3.8-27B yes/no scoring · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish 84.6% (95%: 81.4–87.4%) · English 84.8% (95%: 81.6–87.6%), on the same 547 kept reading questions English minus Turkish: +0.2 points | 84.6%84.8% | 84.6% | 84.8% | +0.2 | |
| Qwen3.8-27B stated confidence · reasoning off · chat Open · 27Bother setting of Qwen3.8-27B Qwen3.8-27B stated confidence · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish 86.3% (95%: 83.2–88.9%) · English 84.8% (95%: 81.6–87.6%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 86.3%84.8% | 86.3% | 84.8% | -1.5 | |
| Qwen3.8-27B vote share · reasoning off · chat Open · 27Bother setting of Qwen3.8-27B Qwen3.8-27B vote share · reasoning off · chat · Alibaba · chat model Parameters: 27B dense, as the maker states it (counted from the released files: 27.78B). Turkish 85.9% (95%: 82.8–88.6%) · English 83.9% (95%: 80.6–86.8%), on the same 547 kept reading questions English minus Turkish: -2.0 points | 85.9%83.9% | 85.9% | 83.9% | -2.0 | |
| – | Perplexity Decider 27B v1.2 API · with web Open · 27B Perplexity Decider 27B v1.2 API · with web · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 90.7% (95%: 87.9–92.8%) · English 90.5% (95%: 87.7–92.7%), on the same 547 kept reading questions English minus Turkish: -0.2 points | 90.7%90.5% | 90.7% | 90.5% | -0.2 |
| 36 | Perplexity Decider 27B v1.3 Open · 27B Perplexity Decider 27B v1.3 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 90.7% (95%: 87.9–92.8%) · English 89.6% (95%: 86.7–91.9%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 90.7%89.6% | 90.7% | 89.6% | -1.1 |
| 37 | GLiDE v1.2 Open · 27B GLiDE v1.2 · Fastino · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 90.5% (95%: 87.7–92.7%) · English 90.3% (95%: 87.5–92.5%), on the same 547 kept reading questions English minus Turkish: -0.2 points | 90.5%90.3% | 90.5% | 90.3% | -0.2 |
| 38 | GLM-5.3-Flash reasoning on · chat Open · 320B MoE · 18B active GLM-5.3-Flash reasoning on · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish 90.1% (95%: 87.3–92.4%) · English 89.6% (95%: 86.7–91.9%), on the same 547 kept reading questions English minus Turkish: -0.5 points | 90.1%89.6% | 90.1% | 89.6% | -0.5 |
| GLM-5.3-Flash reasoning off · chat Open · 320B MoE · 18B activeother setting of GLM-5.3-Flash GLM-5.3-Flash reasoning off · chat · Z.ai · chat model Mixture of experts: 320B total, about 18B active per step. Size class uses the total (320B). Turkish 84.3% (95%: 81.0–87.1%) · English 87.2% (95%: 84.1–89.7%), on the same 547 kept reading questions English minus Turkish: +2.9 points | 84.3%87.2% | 84.3% | 87.2% | +2.9 | |
| 39 | Drex Max v1.1 Closed · Efficient (provisional) Drex Max v1.1 · Nace.AI · decision model Parameters: not disclosed by maker Class: Efficient (provisional), by its introductory price, the only one published: $0.03 per 1M input tokens (https://nace.ai/developers, read 2026-11-10) Turkish 90.1% (95%: 87.3–92.4%) · English 88.5% (95%: 85.5–90.9%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 90.1%88.5% | 90.1% | 88.5% | -1.6 |
| 40 | SubQ Snap Preview Closed · Efficient SubQ Snap Preview · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.05 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-10-13) Turkish 89.9% (95%: 87.1–92.2%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: -4.0 points | 89.9%85.9% | 89.9% | 85.9% | -4.0 |
| 41 | Inkling-Small high · chat · Jul 2026 Open · 276B MoE · 12B active Inkling-Small high · chat · Jul 2026 · Thinking Machines · chat model Mixture of experts: 276B total (counted from the released files: 265.96B), about 12B active per step. Size class uses the total (276B). Turkish 89.4% (95%: 86.5–91.7%) · English 88.3% (95%: 85.3–90.7%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 89.4%88.3% | 89.4% | 88.3% | -1.1 |
| 42 | Claude Haiku 5.5 max · chat Closed · Efficient Claude Haiku 5.5 max · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://openrouter.ai/anthropic/claude-haiku-5.5, read 2026-11-10) Turkish 89.4% (95%: 86.5–91.7%) · English 92.7% (95%: 90.2–94.6%), on the same 547 kept reading questions English minus Turkish: +3.3 points | 89.4%92.7% | 89.4% | 92.7% | +3.3 |
| 43 | GLM-5.3 low · chat Open · Workhorse GLM-5.3 low · chat · Z.ai · chat model Parameters: not stated by the maker (a mixture of experts); counted from the released files: 753.33B. Class: Workhorse, by its standard list price: $1.40 per 1M input tokens (https://openrouter.ai/z-ai/glm-5.3, read 2026-10-28) Turkish 89.2% (95%: 86.3–91.5%) · English 87.8% (95%: 84.7–90.2%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 89.2%87.8% | 89.2% | 87.8% | -1.5 |
| 44 | SubQ Snap 1.1 Closed · Efficient SubQ Snap 1.1 · Subquadratic · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.subq.ai/using-the-api/models/, read 2026-11-01) Turkish 89.2% (95%: 86.3–91.5%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: -3.3 points | 89.2%85.9% | 89.2% | 85.9% | -3.3 |
| 45 | Perplexity Decider 27B v1 Open · 27B Perplexity Decider 27B v1 · Perplexity · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 26.09B). Turkish 89.0% (95%: 86.1–91.4%) · English 88.5% (95%: 85.5–90.9%), on the same 547 kept reading questions English minus Turkish: -0.5 points | 89.0%88.5% | 89.0% | 88.5% | -0.5 |
| 46 | Qwen3.8 Flash no reasoning · chat Closed · Efficient Qwen3.8 Flash no reasoning · chat · Alibaba · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.15 per 1M input tokens (https://openrouter.ai/qwen/qwen3.8-flash, read 2026-11-11) Turkish 88.8% (95%: 85.9–91.2%) · English 87.0% (95%: 83.9–89.6%), on the same 547 kept reading questions English minus Turkish: -1.8 points | 88.8%87.0% | 88.8% | 87.0% | -1.8 |
| 47 | d1.2 Closed · Efficient d1.2 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/blog/d1-2, read 2026-11-16) Turkish 88.8% (95%: 85.9–91.2%) · English 89.0% (95%: 86.1–91.4%), on the same 547 kept reading questions English minus Turkish: +0.2 points | 88.8%89.0% | 88.8% | 89.0% | +0.2 |
| 48 | GPT-6 Luna low · chat Closed · Efficient GPT-6 Luna low · chat · OpenAI · chat model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish 88.7% (95%: 85.7–91.1%) · English 86.3% (95%: 83.2–88.9%), on the same 547 kept reading questions English minus Turkish: -2.4 points | 88.7%86.3% | 88.7% | 86.3% | -2.4 |
| 49 | d1.1 Closed · Efficient d1.1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-24) Turkish 88.3% (95%: 85.3–90.7%) · English 88.3% (95%: 85.3–90.7%), on the same 547 kept reading questions English minus Turkish: +0.0 points | 88.3%88.3% | 88.3% | 88.3% | +0.0 |
| 50 | Clef Open · 27B Clef · Cloudflare · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.48B). Turkish 88.1% (95%: 85.1–90.6%) · English 86.7% (95%: 83.5–89.3%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 88.1%86.7% | 88.1% | 86.7% | -1.5 |
| 51 | OneJev-27B Open · 27B OneJev-27B · OmniJev · decision model Parameters: 27B dense, as the maker states it (counted from the released files: 27.36B). Turkish 87.8% (95%: 84.7–90.2%) · English 86.3% (95%: 83.2–88.9%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 87.8%86.3% | 87.8% | 86.3% | -1.5 |
| 52 | Claude Sonnet 5.5 high · chat Closed · Workhorse Claude Sonnet 5.5 high · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 87.6% (95%: 84.5–90.1%) · English 86.1% (95%: 83.0–88.8%), on the same 547 kept reading questions English minus Turkish: -1.5 points | 87.6%86.1% | 87.6% | 86.1% | -1.5 |
| Claude Sonnet 5.5 high · chat · 1 per request Closed · Workhorseother setting of Claude Sonnet 5.5 Claude Sonnet 5.5 high · chat · 1 per request · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 94.3% (95%: 92.1–96.0%) · English 91.6% (95%: 89.0–93.6%), on the same 547 kept reading questions English minus Turkish: -2.7 points | 94.3%91.6% | 94.3% | 91.6% | -2.7 | |
| Claude Sonnet 5.5 low · chat Closed · Workhorseother setting of Claude Sonnet 5.5 Claude Sonnet 5.5 low · chat · Anthropic · chat model Parameters: not disclosed by maker Class: Workhorse, by its standard list price: $2.00 per 1M input tokens (https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-08) Turkish 84.8% (95%: 81.6–87.6%) · English 83.9% (95%: 80.6–86.8%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 84.8%83.9% | 84.8% | 83.9% | -0.9 | |
| 53 | KnowLine-12B-Gen1 Open · 12B KnowLine-12B-Gen1 · PelaAI · decision model Base model: Gemma 4 12B Unified Parameters: 12B dense, as the maker states it (counted from the released files: 11.96B). Turkish 87.6% (95%: 84.5–90.1%) · English 86.7% (95%: 83.5–89.3%), on the same 547 kept reading questions English minus Turkish: -0.9 points | 87.6%86.7% | 87.6% | 86.7% | -0.9 |
| 54 | celeris-1-decision Closed · Efficient celeris-1-decision · Celeris · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://docs.celeris.ai/pricing, read 2026-10-10) Turkish 87.4% (95%: 84.3–89.9%) · English 84.1% (95%: 80.8–86.9%), on the same 547 kept reading questions English minus Turkish: -3.3 points | 87.4%84.1% | 87.4% | 84.1% | -3.3 |
| 55 | Jev 1.14 Closed · Efficient Jev 1.14 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://docs.typesafe.ai/models, read 2026-10-16) Turkish 87.2% (95%: 84.1–89.7%) · English 86.1% (95%: 83.0–88.8%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 87.2%86.1% | 87.2% | 86.1% | -1.1 |
| 56 | Mercury Decide Closed · Efficient Mercury Decide · Inception · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://www.inceptionlabs.ai/, read 2026-11-10) Turkish 87.0% (95%: 83.9–89.6%) · English 85.7% (95%: 82.6–88.4%), on the same 547 kept reading questions English minus Turkish: -1.3 points | 87.0%85.7% | 87.0% | 85.7% | -1.3 |
| 57 | MiMo-V2.6-Flash high · chat Open · 309B MoE · 15B active MiMo-V2.6-Flash high · chat · Xiaomi · chat model Mixture of experts: 309B total, about 15B active per step. Size class uses the total (309B). Turkish 86.8% (95%: 83.7–89.4%) · English 89.0% (95%: 86.1–91.4%), on the same 547 kept reading questions English minus Turkish: +2.2 points | 86.8%89.0% | 86.8% | 89.0% | +2.2 |
| 58 | GPT-6 Luna Decisions API Closed · Efficient GPT-6 Luna Decisions API · OpenAI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.10 per 1M input tokens (https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability, read 2026-10-08) Turkish 86.7% (95%: 83.5–89.3%) · English 84.1% (95%: 80.8–86.9%), on the same 547 kept reading questions English minus Turkish: -2.6 points | 86.7%84.1% | 86.7% | 84.1% | -2.6 |
| 59 | Open-Jev-27B-v1.1 Open · 27B Open-Jev-27B-v1.1 · Zefan Cai · decision model Base model: Qwen3.8-27B + adapter Parameters: 27B dense, as the maker states it (counted from the released files: 27.80B). Turkish 86.5% (95%: 83.3–89.1%) · English 85.4% (95%: 82.2–88.1%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 86.5%85.4% | 86.5% | 85.4% | -1.1 |
| 60 | Jeffan-9B-1211 Open · 9.0B Jeffan-9B-1211 · tattrongvu · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish 86.5% (95%: 83.3–89.1%) · English 87.6% (95%: 84.5–90.1%), on the same 547 kept reading questions English minus Turkish: +1.1 points | 86.5%87.6% | 86.5% | 87.6% | +1.1 |
| 61 | MiniMax-M3.1-Flash high · chat Open · 428B MoE · 23B active MiniMax-M3.1-Flash high · chat · MiniMax · chat model Mixture of experts: 428B total, about 23B active per step. Size class uses the total (428B). Turkish 86.3% (95%: 83.2–88.9%) · English 84.5% (95%: 81.2–87.3%), on the same 547 kept reading questions English minus Turkish: -1.8 points | 86.3%84.5% | 86.3% | 84.5% | -1.8 |
| 62 | Jev 1.13 Closed · Efficient Jev 1.13 · TypeSafe · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://typesafe.ai/blog/introducing-system-one-models-and-jev, read 2026-10-08) Turkish 86.1% (95%: 83.0–88.8%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: -0.2 points | 86.1%85.9% | 86.1% | 85.9% | -0.2 |
| 63 | Cygnet Open · 12B Cygnet · Blockbrain Labs · decision model Base model: Gemma 4 12B IT Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.96B). Turkish 85.6% (95%: 82.4–88.3%) · English 83.9% (95%: 80.6–86.8%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 85.6%83.9% | 85.6% | 83.9% | -1.6 |
| 64 | d1 Closed · Efficient d1 · Liquid AI · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.02 per 1M input tokens (https://www.liquid.ai/pricing, read 2026-10-18) Turkish 85.0% (95%: 81.8–87.8%) · English 86.8% (95%: 83.7–89.4%), on the same 547 kept reading questions English minus Turkish: +1.8 points | 85.0%86.8% | 85.0% | 86.8% | +1.8 |
| 65 | Winnow-12B Open · 12B Winnow-12B · EldanRing · decision model Parameters: 11.95B dense, as the maker states it (counted from the released files: 11.91B). Turkish 85.0% (95%: 81.8–87.8%) · English 83.9% (95%: 80.6–86.8%), on the same 547 kept reading questions English minus Turkish: -1.1 points | 85.0%83.9% | 85.0% | 83.9% | -1.1 |
| 66 | Drex 2.0 Open · 35B MoE · 3.0B active Drex 2.0 · Nace.AI · decision model Mixture of experts: 35B total (counted from the released files: 35.11B), about 3B active per step. Size class uses the total (35B). Turkish 84.8% (95%: 81.6–87.6%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: +1.1 points | 84.8%85.9% | 84.8% | 85.9% | +1.1 |
| 67 | GLiDE 9B Open · 7.9B GLiDE 9B · Fastino · decision model Parameters: 7.94B dense, as the maker states it (counted from the released files: 7.94B). Turkish 84.3% (95%: 81.0–87.1%) · English 86.3% (95%: 83.2–88.9%), on the same 547 kept reading questions English minus Turkish: +2.0 points | 84.3%86.3% | 84.3% | 86.3% | +2.0 |
| 68 | Adjudex 9B v5 Open · 9.0B Adjudex 9B v5 · newsbang · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish 83.9% (95%: 80.6–86.8%) · English 85.7% (95%: 82.6–88.4%), on the same 547 kept reading questions English minus Turkish: +1.8 points | 83.9%85.7% | 83.9% | 85.7% | +1.8 |
| 69 | Elephant-1 Open · 9.0B Elephant-1 · Base Labs · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.41B). Turkish 83.5% (95%: 80.2–86.4%) · English 85.4% (95%: 82.2–88.1%), on the same 547 kept reading questions English minus Turkish: +1.8 points | 83.5%85.4% | 83.5% | 85.4% | +1.8 |
| 70 | Cerno 9B v1.2 Open · 9.0B Cerno 9B v1.2 · Apertix AI · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 8.95B). Turkish 83.5% (95%: 80.2–86.4%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: +2.4 points | 83.5%85.9% | 83.5% | 85.9% | +2.4 |
| 71 | Dexy-Decision-9B-V2 Open · 9.0B Dexy-Decision-9B-V2 · Patrick-18 · decision model Base model: vllm-sr/d3-flash Parameters: 9B dense, as the maker states it (counted from the released files: 8.39B). Turkish 83.4% (95%: 80.0–86.2%) · English 83.5% (95%: 80.2–86.4%), on the same 547 kept reading questions English minus Turkish: +0.2 points | 83.4%83.5% | 83.4% | 83.5% | +0.2 |
| 72 | StartLux-Decision-v1.1-9B-Instant Open · 9.0B StartLux-Decision-v1.1-9B-Instant · StartLux · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish 83.0% (95%: 79.6–85.9%) · English 86.8% (95%: 83.7–89.4%), on the same 547 kept reading questions English minus Turkish: +3.8 points | 83.0%86.8% | 83.0% | 86.8% | +3.8 |
| 73 | Qwen3.5-4B reasoning on · chat Open · 4.0B Qwen3.5-4B reasoning on · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish 82.8% (95%: 79.4–85.7%) · English 87.0% (95%: 83.9–89.6%), on the same 547 kept reading questions English minus Turkish: +4.2 points | 82.8%87.0% | 82.8% | 87.0% | +4.2 |
| 74 | KnowLine-9B-Gen2 Open · 9.0B KnowLine-9B-Gen2 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish 82.8% (95%: 79.4–85.7%) · English 85.2% (95%: 82.0–87.9%), on the same 547 kept reading questions English minus Turkish: +2.4 points | 82.8%85.2% | 82.8% | 85.2% | +2.4 |
| 75 | KnowLine-9B-Gen1 Open · 9.0B KnowLine-9B-Gen1 · PelaAI · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 9.65B). Turkish 82.4% (95%: 79.0–85.4%) · English 85.9% (95%: 82.8–88.6%), on the same 547 kept reading questions English minus Turkish: +3.5 points | 82.4%85.9% | 82.4% | 85.9% | +3.5 |
| 76 | Gemma 4 E4B default reasoning · chat Open · 8.0B Gemma 4 E4B default reasoning · chat · Google · chat model Parameters: 8B dense · 4.5B effective, as the maker states it (counted from the released files: 8.00B). Turkish 80.6% (95%: 77.1–83.7%) · English 82.4% (95%: 79.0–85.4%), on the same 547 kept reading questions English minus Turkish: +1.8 points | 80.6%82.4% | 80.6% | 82.4% | +1.8 |
| 77 | vera-medium-omni Open · 4.2B vera-medium-omni · scar-ai · decision model Base model: Qwen3.5-4B Parameters: 4.2B dense, as the maker states it (counted from the released files: 4.21B). Turkish 79.5% (95%: 75.9–82.7%) · English 83.5% (95%: 80.2–86.4%), on the same 547 kept reading questions English minus Turkish: +4.0 points | 79.5%83.5% | 79.5% | 83.5% | +4.0 |
| 78 | Valen-Decision-1.5-9B Open · 9.0B Valen-Decision-1.5-9B · Valen-Team · decision model Base model: Qwen3.5-9B Parameters: 9B dense, as the maker states it (counted from the released files: 8.40B). Turkish 79.0% (95%: 75.4–82.2%) · English 83.7% (95%: 80.4–86.6%), on the same 547 kept reading questions English minus Turkish: +4.8 points | 79.0%83.7% | 79.0% | 83.7% | +4.8 |
| 79 | Perplexity Decider Flash v1.3 Open · 4.0B Perplexity Decider Flash v1.3 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish 78.8% (95%: 75.2–82.0%) · English 83.7% (95%: 80.4–86.6%), on the same 547 kept reading questions English minus Turkish: +4.9 points | 78.8%83.7% | 78.8% | 83.7% | +4.9 |
| 80 | KnowLine 4B Gen5.5 Open · 4.0B KnowLine 4B Gen5.5 · PelaAI · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish 78.6% (95%: 75.0–81.8%) · English 80.8% (95%: 77.3–83.9%), on the same 547 kept reading questions English minus Turkish: +2.2 points | 78.6%80.8% | 78.6% | 80.8% | +2.2 |
| 81 | StartLux-Decision-v1.1-4B-Instant Open · 4.0B StartLux-Decision-v1.1-4B-Instant · StartLux · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish 78.4% (95%: 74.8–81.7%) · English 82.8% (95%: 79.4–85.7%), on the same 547 kept reading questions English minus Turkish: +4.4 points | 78.4%82.8% | 78.4% | 82.8% | +4.4 |
| 82 | Clef-Flash Open · 9.0B Clef-Flash · Cloudflare · decision model Parameters: 9B dense, as the maker states it (counted from the released files: 9.53B). Turkish 77.7% (95%: 74.0–81.0%) · English 82.3% (95%: 78.8–85.2%), on the same 547 kept reading questions English minus Turkish: +4.6 points | 77.7%82.3% | 77.7% | 82.3% | +4.6 |
| 83 | Verdin-4B-v1.1 Open · 4.0B Verdin-4B-v1.1 · victorbennett · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish 77.5% (95%: 73.8–80.8%) · English 82.3% (95%: 78.8–85.2%), on the same 547 kept reading questions English minus Turkish: +4.8 points | 77.5%82.3% | 77.5% | 82.3% | +4.8 |
| 84 | Qwen3-14B reasoning off · chat Open · 15B Qwen3-14B reasoning off · chat · Alibaba · chat model Parameters: 14.8B dense, as the maker states it (counted from the released files: 14.77B). Turkish 76.8% (95%: 73.1–80.1%) · English 79.7% (95%: 76.1–82.9%), on the same 547 kept reading questions English minus Turkish: +2.9 points | 76.8%79.7% | 76.8% | 79.7% | +2.9 |
| 85 | Perplexity Decider Flash v1.2 Open · 4.0B Perplexity Decider Flash v1.2 · Perplexity · decision model Parameters: 4B dense, as the maker states it (counted from the released files: 4.54B). Turkish 76.6% (95%: 72.9–80.0%) · English 81.9% (95%: 78.5–84.9%), on the same 547 kept reading questions English minus Turkish: +5.3 points | 76.6%81.9% | 76.6% | 81.9% | +5.3 |
| 86 | Microsoft-Decision-1 Closed · Efficient Microsoft-Decision-1 · Microsoft · decision model Base model: Qwen3.5-9B Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.042 per 1M input tokens (https://openrouter.ai/microsoft/microsoft-decision-1, read 2026-10-17) Turkish 76.4% (95%: 72.7–79.8%) · English 85.7% (95%: 82.6–88.4%), on the same 547 kept reading questions English minus Turkish: +9.3 points | 76.4%85.7% | 76.4% | 85.7% | +9.3 |
| 87 | Chinese Decision Pointer 4B Open · 4.0B Chinese Decision Pointer 4B · quincyqiang · decision model Base model: Qwen3.5-4B Parameters: 4B dense, as the maker states it (counted from the released files: 4.21B). Turkish 75.7% (95%: 71.9–79.1%) · English 80.4% (95%: 76.9–83.5%), on the same 547 kept reading questions English minus Turkish: +4.8 points | 75.7%80.4% | 75.7% | 80.4% | +4.8 |
| 88 | Erk-14B reasoning off · chat Open · 14B Erk-14B reasoning off · chat · eCloud Technology · chat model Parameters: 14B dense, as the maker states it (counted from the released files: 14.77B). Turkish 75.3% (95%: 71.5–78.7%) · English 72.0% (95%: 68.1–75.6%), on the same 547 kept reading questions English minus Turkish: -3.3 points | 75.3%72.0% | 75.3% | 72.0% | -3.3 |
| 89 | Qwen3.5-4B reasoning off · chat Open · 4.0B Qwen3.5-4B reasoning off · chat · Alibaba · chat model Parameters: 4B dense, as the maker states it (counted from the released files: 4.66B). Turkish 72.0% (95%: 68.1–75.6%) · English 80.4% (95%: 76.9–83.5%), on the same 547 kept reading questions English minus Turkish: +8.4 points | 72.0%80.4% | 72.0% | 80.4% | +8.4 |
| 90 | Metask-Jev-4B Open · 4.5B Metask-Jev-4B · Metask Lab · decision model Parameters: 4.54B dense, as the maker states it (counted from the released files: 4.54B). Turkish 67.1% (95%: 63.0–70.9%) · English 77.1% (95%: 73.4–80.5%), on the same 547 kept reading questions English minus Turkish: +10.1 points | 67.1%77.1% | 67.1% | 77.1% | +10.1 |
| 91 | KnowLine-2B-Gen1 Open · 2.0B KnowLine-2B-Gen1 · PelaAI · decision model Base model: Qwen3.5-2B Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish 58.0% (95%: 53.8–62.0%) · English 71.3% (95%: 67.4–74.9%), on the same 547 kept reading questions English minus Turkish: +13.3 points | 58.0%71.3% | 58.0% | 71.3% | +13.3 |
| 92 | StartLux-Decision-v1.1-2B-Instant Open · 2.0B StartLux-Decision-v1.1-2B-Instant · StartLux · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish 55.8% (95%: 51.6–59.9%) · English 69.7% (95%: 65.7–73.4%), on the same 547 kept reading questions English minus Turkish: +13.9 points | 55.8%69.7% | 55.8% | 69.7% | +13.9 |
| 93 | GLiDE 2B Open · 1.9B GLiDE 2B · Fastino · decision model Parameters: 1.88B dense, as the maker states it (counted from the released files: 1.88B). Turkish 54.3% (95%: 50.1–58.4%) · English 70.0% (95%: 66.0–73.7%), on the same 547 kept reading questions English minus Turkish: +15.7 points | 54.3%70.0% | 54.3% | 70.0% | +15.7 |
| 94 | LFM3-1.7B reasoning on · chat Open · 1.7B LFM3-1.7B reasoning on · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish 48.1% (95%: 43.9–52.3%) · English 71.1% (95%: 67.2–74.8%) (historical paired baseline, run 1061), on the same 547 kept reading questions English minus Turkish: +23.0 points | 48.1%71.1% | 48.1% | 71.1%historical paired baseline, run 1061 | +23.0 |
| 95 | LFM2.5-2.6B default reasoning · chat Open · 2.7B LFM2.5-2.6B default reasoning · chat · Liquid AI · chat model Parameters: 2.69B dense, as the maker states it (counted from the released files: 2.70B). Turkish 47.9% (95%: 43.7–52.1%) · English 73.9% (95%: 70.0–77.4%), on the same 547 kept reading questions English minus Turkish: +26.0 points | 47.9%73.9% | 47.9% | 73.9% | +26.0 |
| 96 | GLiDE 0.8B Open · 0.8B GLiDE 0.8B · Fastino · decision model Parameters: 750M dense, as the maker states it (counted from the released files: 0.75B). Turkish 46.1% (95%: 41.9–50.3%) · English 60.7% (95%: 56.5–64.7%), on the same 547 kept reading questions English minus Turkish: +14.6 points | 46.1%60.7% | 46.1% | 60.7% | +14.6 |
| 97 | OneJev1.6-0.8B Open · 0.8B OneJev1.6-0.8B · OmniJev · decision model Parameters: 800M dense, as the maker states it (counted from the released files: 0.85B). Turkish 42.8% (95%: 38.7–47.0%) · English 59.4% (95%: 55.2–63.5%), on the same 547 kept reading questions English minus Turkish: +16.6 points | 42.8%59.4% | 42.8% | 59.4% | +16.6 |
| 98 | d1-2B Open · 1.9B d1-2B · Liquid AI · decision model Base model: LFM3-VL-2B Parameters: 1.92B dense, as the maker states it (counted from the released files: 1.92B). Turkish 41.9% (95%: 37.8–46.0%) · English 65.6% (95%: 61.6–69.5%), on the same 547 kept reading questions English minus Turkish: +23.8 points | 41.9%65.6% | 41.9% | 65.6% | +23.8 |
| 99 | KnowLine-0.8B-Gen1 Open · 0.8B KnowLine-0.8B-Gen1 · PelaAI · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish 40.2% (95%: 36.2–44.4%) · English 58.9% (95%: 54.7–62.9%), on the same 547 kept reading questions English minus Turkish: +18.6 points | 40.2%58.9% | 40.2% | 58.9% | +18.6 |
| 100 | LFM3-1.7B reasoning off · chat Open · 1.7B LFM3-1.7B reasoning off · chat · Liquid AI · chat model Parameters: 1.73B dense, as the maker states it (counted from the released files: 1.73B). Turkish 40.2% (95%: 36.2–44.4%) · English 72.9% (95%: 69.1–76.5%) (historical paired baseline, run 1223), on the same 547 kept reading questions English minus Turkish: +32.7 points | 40.2%72.9% | 40.2% | 72.9%historical paired baseline, run 1223 | +32.7 |
| 101 | Strands Decider 2B Open · 2.0B Strands Decider 2B · Strands Agents · decision model Parameters: 2B dense, as the maker states it (counted from the released files: 2.29B). Turkish 39.1% (95%: 35.1–43.3%) · English 45.5% (95%: 41.4–49.7%), on the same 547 kept reading questions English minus Turkish: +6.4 points | 39.1%45.5% | 39.1% | 45.5% | +6.4 |
| 102 | StartLux-Decision-v1.1-0.8B-Instant Open · 0.8B StartLux-Decision-v1.1-0.8B-Instant · StartLux · decision model Base model: Qwen3.5-0.8B Parameters: 800M dense, as the maker states it (counted from the released files: 0.87B). Turkish 37.3% (95%: 33.3–41.4%) · English 59.6% (95%: 55.4–63.6%), on the same 547 kept reading questions English minus Turkish: +22.3 points | 37.3%59.6% | 37.3% | 59.6% | +22.3 |
| 103 | Qwen3.5-2B reasoning off · chat Open · 2.0B Qwen3.5-2B reasoning off · chat · Alibaba · chat model Parameters: 2B dense, as the maker states it (counted from the released files: 2.27B). Turkish 36.7% (95%: 32.8–40.9%) · English 67.3% (95%: 63.2–71.1%), on the same 547 kept reading questions English minus Turkish: +30.5 points | 36.7%67.3% | 36.7% | 67.3% | +30.5 |
| 104 | Decider 1 Closed · Efficient Decider 1 · meraGPT · decision model Parameters: not disclosed by maker Class: Efficient, by its standard list price: $0.04 per 1M input tokens (https://meragpt.com/docs/models, read 2026-10-19) Turkish 36.6% (95%: 32.6–40.7%) · English 55.8% (95%: 51.6–59.9%), on the same 547 kept reading questions English minus Turkish: +19.2 points | 36.6%55.8% | 36.6% | 55.8% | +19.2 |
| 105 | LFM2.5-8B-A1B default reasoning · chat Open · 8.3B MoE · 1.5B active LFM2.5-8B-A1B default reasoning · chat · Liquid AI · chat model Mixture of experts: 8.3B total (counted from the released files: 8.47B), about 1.5B active per step. Size class uses the total (8.3B). Turkish 27.1% (95%: 23.5–30.9%) · English 70.4% (95%: 66.4–74.1%), on the same 547 kept reading questions English minus Turkish: +43.3 points | 27.1%70.4% | 27.1% | 70.4% | +43.3 |
| 106 | GLiClass v4 Edge Open · 0.4B GLiClass v4 Edge · Knowledgator · decision model Base model: DeBERTa-v3-large Parameters: 400M dense, as the maker states it (counted from the released files: 0.44B). Turkish 26.9% (95%: 23.3–30.7%) · English 58.3% (95%: 54.1–62.4%), on the same 547 kept reading questions English minus Turkish: +31.4 points | 26.9%58.3% | 26.9% | 58.3% | +31.4 |
| 107 | GLiDE 0.4B Open · 0.4B GLiDE 0.4B · Fastino · decision model Parameters: 400M dense, as the maker states it (counted from the released files: 0.40B). Turkish 25.8% (95%: 22.3–29.6%) · English 59.4% (95%: 55.2–63.5%), on the same 547 kept reading questions English minus Turkish: +33.6 points | 25.8%59.4% | 25.8% | 59.4% | +33.6 |
| 108 | GLiNER2.5-multi-Decide Open · 0.3B GLiNER2.5-multi-Decide · Fastino · decision model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish 20.7% (95%: 17.5–24.2%) · English 21.2% (95%: 18.0–24.8%), on the same 547 kept reading questions English minus Turkish: +0.5 points | 20.7%21.2% | 20.7% | 21.2% | +0.5 |
| 109 | Laya Open · 0.4B Laya · Convai Innovations · decision model Parameters: 421M dense, as the maker states it (counted from the released files: 0.42B). Turkish 20.5% (95%: 17.3–24.1%) · English 22.9% (95%: 19.5–26.6%), on the same 547 kept reading questions English minus Turkish: +2.4 points | 20.5%22.9% | 20.5% | 22.9% | +2.4 |
| 110 | Laya Multilingual Open · 0.3B Laya Multilingual · Convai Innovations · decision model Parameters: 322M dense, as the maker states it (counted from the released files: 0.32B). Turkish 17.0% (95%: 14.1–20.4%) · English 15.4% (95%: 12.6–18.6%), on the same 547 kept reading questions English minus Turkish: -1.6 points | 17.0%15.4% | 17.0% | 15.4% | -1.6 |
| 111 | GLiNER2.5 Multi Open · 0.3B GLiNER2.5 Multi · Fastino · chat model Parameters: 287M dense, as the maker states it (counted from the released files: 0.29B). Turkish 16.8% (95%: 13.9–20.2%) · English 16.3% (95%: 13.4–19.6%), on the same 547 kept reading questions English minus Turkish: -0.5 points | 16.8%16.3% | 16.8% | 16.3% | -0.5 |
| 112 | MiniCPM5-2B reasoning off · chat Open · 2.5B MiniCPM5-2B reasoning off · chat · OpenBMB · chat model Parameters: 2.52B dense, as the maker states it. Turkish 7.1% (95%: 5.3–9.6%) · English 67.6% (95%: 63.6–71.4%) (historical paired baseline, run 1094), on the same 547 kept reading questions English minus Turkish: +60.5 points | 7.1%67.6% | 7.1% | 67.6%historical paired baseline, run 1094 | +60.5 |
Turkish (solid) and English (striped) accuracy on the same 547 kept reading questions, with 95% intervals, ranked by Turkish, as on English and Turkish.
No row of this part matches these keys. The small open chat models ran in English only.
Rank: the row's place among the rows in view, by accuracy, 1, 2, 3 without repeats; rows with the same score keep the table's order, which claims no difference between them. No all-pairs test covers the Both part, so it has no tiers. A setup with web search is not ranked (Rank: –).
One row per model at its highest reasoning setting, never its best score; tap or point at a row for its details. How rows are chosen
Size by total parameters, as disclosed by the maker (for MoE models the label also shows active parameters): Tiny under 6B · Small 6B–15B · Medium over 15B up to 40B · Large over 40B. Models of undisclosed size, by standard list price per 1M input tokens: Efficient up to $0.50 · Workhorse up to $3 · Flagship over $3. Provisional: Drex Max v1.1, whose maker publishes only a promotional price. Left out here: the setups only this site ran (Erk-14B, GLiNER2.5 Multi, GLiNER2.5-multi-Decide, Laya, Laya Multilingual and Qwen3-14B); they show under Class: All.
In progress
Models joining Reading (the project lead via Zeki #5064, #5071; Ümit #5067). Each side shows its state until its run is published; nothing here is a score. As of 15 November 2026, 14:05 UTC.
| Model | English | Turkish | Page images |
|---|---|---|---|
| Gemma 4 E4BGoogle · chat | Done | Done | Queuedpicture only, text plus picture and two OCR routes done; the other two in the Turkish phase (Zeki #5213) |
| MiniCPM5-2BOpenBMB · chat | Done | SkippedÜmit chose not to run it after its Turkish smoke failed (run 1120: 15 of 20 replies without a final answer line; Zeki #7178, #8830) | – |
| Inkling 2 SmallThinking Machines · chat | Runningthrough OpenRouter (Nebius), started 16 November 2026 (Ümit #9817, #10509) | Runningthrough OpenRouter (Nebius) | – |
| Spark-X2.6-4BXHToken · chat | Queued | Queued | – |
| Qwen3.5-4BAlibaba · chat | Done | Done | Queuedin the Turkish phase (Zeki #5213) |
| Gemma 4 E2BGoogle · chat | Done | Queuedin logic's GPU 1 queue on a load of its own (thinking off, 8,192 tokens), after Granite (Ümit #8343, logic #8391) | Queuedin the Turkish phase (Zeki #5213) |
| Gemma 4 12B UnifiedGoogle · chat | Done | Queuedin logic's GPU 1 queue, on its 32,768-token setup's server, which also answers to this setup's name (Ümit #6527, logic #6955) | Queuedin the Turkish phase (Zeki #5213) |
| Granite-4.2-3BIBM · chat | Done | QueuedText stopped at 741 of 2,198 so that other setups go first; it resumes on logic's next Granite load, then "Bilmiyorum" (logic #7481, #7599) | – |
| Qwen3.5-9BAlibaba · chat | Queued | Queued | Queued |
| Granite-4.2-8BIBM · chat | Queued | Queued | – |
| LFM2.5-24B-A2BLiquid AI · chat | Done | Queuednot scheduled yet: needs the model served again on logic's GPU 1 (Ümit #6527) | – |
| Granite-4.2-30BIBM · chat | Queued | Queued | – |
| Granite-4.2-3B (thinking off)IBM · chat | Queuedin logic's GPU 1 queue (Ümit #6924, logic #6955) | Queuedwith its English runs, same server | – |
| Qwen3.5-2BAlibaba · chat | Done | Queuednot asked yet: Ümit's grid asked for the English runs (#10884) | – |
| Gemma 4 E2B (thinking on, 32,768 tokens)Google · chat | Done | Queuednot asked yet: Ümit's grid asked for the English runs (#10884) | – |
| Gemma 4 E4B (thinking on, 32,768 tokens)Google · chat | Queuedin logic's GPU 1 queue after the thinking-off setups (Ümit #6523, logic #6955) | Queuedwith its English runs, same server | – |
| Gemma 4 12B Unified (thinking on, 32,768 tokens)Google · chat | Queuedlast in logic's GPU 1 queue (Ümit #6924, logic #6955) | Queuedwith its English runs, same server | – |
| Gemma 4 31B (thinking on)Google · chat | Queuednot set up yet: another server, after logic's GPU 1 queue (Ümit #6924, logic #6929) | Queuedwith its English runs | – |
| Qwen3-14B (thinking on)Alibaba · chat | Queuednot set up yet: another server, after logic's GPU 1 queue (Ümit #6924, logic #6929) | Queuedwith its English runs | – |
| Erk-14B (thinking on)eCloud Technology · chat | Queuednot set up yet: another server, after logic's GPU 1 queue (Ümit #6924, logic #6929) | Queuedwith its English runs | – |
| Raon-1 nanoKRAFTON · chat | Runningthinking on, 32,768 tokens, on KRAFTON's own vLLM engine on logic's GPU 1 since 19:01 UTC (Ümit #10884) | Queuednot asked yet: Ümit's grid asked for the English runs (#10884) | – |
| Raon-1 nano (thinking off)KRAFTON · chat | Done | Queuedasked on its next load in logic's GPU 1 queue, whichever way its smoke came out (run 1154: 3 of 20 replies at the 32,768-token limit, 5 without a letter; Ümit #6515 via Zeki #8211) | – |
| d1-Omni-2BLiquid AI · decision | Queued | Queued | Queued |
| d1-3BLiquid AI · decision | Queued | Queued | Queued |
| KnowLine-2B-Gen1PelaAI · decision | Done | Done | Queuedin the Turkish phase (Zeki #5213) |
| Sieve 31B v3Sthānika AI · decision | Done | Done | Queuedthe four OCR routes only (its server reads text), in the Turkish phase (Zeki #5213) |
| Turion Jev 31BTurion.AI · decision | Done | Done | Queuedthe four OCR routes only (its server reads text), in the Turkish phase (Zeki #5213) |
| StartLux-Decision-v1.1-4B-InstantStartLux · decision | Done | Done | Queuedin the Turkish phase (Zeki #5213) |
| Cerno 9B v1.2Apertix AI · decision | Done | Done | Queuedthe four OCR routes only (its server reads text), in the Turkish phase (Zeki #5213) |
| Muse Glimmer 30BMeta · chat | Queued | Queued | Queued |
| Muse EmberMeta · chat | Done | Done | Queuedfive cells done; the last OCR route paused at 562 of 763, the rest in the Turkish phase (Zeki #5213, #5228) |
Chat models run on logic's server on bariton GPU 1 after its Logic & Math runs, one model at a time (logic #5080, #5091, #5214); since 14 November logic's queue runs Reading's arms on the same server right after its own (logic #6929). English came first for every model (the project lead via Zeki #5213); for the small chat models Ümit asked for the Turkish runs too (#6224, #6527) and for a thinking-off and a thinking-on setup of each model that has the switch (#6924), each setup a row of its own.
Page images: chat models that read images take the same conditions through the chat route on their own server, one question per request with the plain prompt (accepted by Zeki #5157); – means the model reads no images.
Muse Glimmer 30B, d1-Omni-2B and d1-3B are being set up (logic #5089); their runs follow the brief's phases (Ümit #5067, Zeki #5090).
All setups: the full ranking
Swipe for more columns →
All models
| Model | Accuracy (%) | Reading (%) | Grammar (%) | Median latency per request (s) | $ per 1,000 questions |
|---|---|---|---|---|---|
| Gemini 4 Carbon high · chat | 98.597.8 to 98.9 | 97.6 | 98.8 | 217.7 | 31.309 |
| Claude Fable 5.2 max · chat | 98.197.5 to 98.6 | 97.7 | 98.3 | 81.1 | 53.381 |
| Gemini 3.8 Flash high · chat | 98.197.4 to 98.6 | 97.1 | 98.5 | 53.7 | 2.419 |
| Gemini 4 Hydrogen high · chat | 98.197.4 to 98.6 | 97.3 | 98.4 | 31.7 | 4.806 |
| GPT-6 Astra high · chat | 98.097.3 to 98.5 | 97.1 | 98.3 | 21.7 | 6.066 |
| GPT-6.1 Sol high · chat | 97.897.1 to 98.3 | 96.5 | 98.3 | 24.7 | 1.622 |
| GPT-6 Astra low · chat | 97.696.9 to 98.2 | 97.0 | 97.9 | 13.0* | 5.378 |
| GPT-6 Astra medium · chat | 97.596.8 to 98.1 | 97.0 | 97.7 | 15.3 | 5.447 |
| GPT-6 Astra low · chat · 1 per request | 97.496.7 to 98.0 | 97.3 | 97.5 | 6.41 | 26.738 |
| Gemini 3.8 Flash low · chat · 1 per request | 97.196.3 to 97.8 | 96.2 | 97.5 | 2.72 | 0.963 |
| GPT-6.1 Sol low · chat · 1 per request | 96.795.9 to 97.4 | 96.0 | 97.0 | 6.61 | 5.569 |
| GPT-6.1 Sol low · chat | 96.795.8 to 97.4 | 96.2 | 96.9 | 13.3 | 1.130 |
| Gemini 4 Argon low · chat | 96.495.5 to 97.1 | 97.0 | 96.1 | 5.46 | 1.786 |
| Claude Fable 5.1 high · chat | 96.195.2 to 96.9 | 95.6 | 96.4 | 30.3 | 17.042 |
| Claude Opus 5.5 high · chat | 95.094.0 to 95.8 | 95.6 | 94.7 | 22.1 | 6.509 |
| GPT-6.1 Terra high · chat | 94.393.2 to 95.2 | 96.2 | 93.5 | 7.10 | 0.515 |
| Claude Opus 5.5 low · chat · 1 per request | 93.392.2 to 94.3 | 95.7 | 92.3 | 3.71 | 11.610 |
| Gemini 3.8 Flash low · chat | 93.392.1 to 94.2 | 95.1 | 92.5 | 2.19 | 0.269 |
| Grok 4.7 high · chat | 92.791.6 to 93.7 | 95.0 | 91.8 | 401.9 | 14.292 |
| DeepSeek-V4.1-Flash reasoning on · chat · first release | 92.090.8 to 93.1 | 93.4 | 91.4 | 365.1 | 1.299 |
| DeepSeek-V4.1-Flash reasoning on · chat · 1017 | 91.890.5 to 92.8 | 94.1 | 90.8 | 486.9 | 1.278 |
| Qwen4-Max high · chat | 91.790.5 to 92.8 | 94.7 | 90.5 | 91.5 | 2.533 |
| GPT-6.1 Terra low · chat | 90.989.6 to 92.0 | 95.9 | 88.7 | 4.24 | 0.426 |
| Claude Opus 5.5 low · chat | 90.489.1 to 91.6 | 94.8 | 88.6 | 13.4* | 4.931 |
| Claude Sonnet 5.5 high · chat · 1 per request | 89.287.8 to 90.4 | 94.2 | 87.0 | 4.64 | 7.748 |
| Muse Ember max · chat | 85.684.1 to 87.0 | 90.2 | 83.7 | 30.9 | 0.650 |
| Claude Sonnet 5.5 high · chat | 85.083.5 to 86.5 | 85.2 | 85.0 | 20.5* | 3.654 |
| Kimi K3.1 Pro low · chat | 82.881.1 to 84.3 | 93.8 | 78.1 | 44.9 | 2.048 |
| Qwen3.5-397B-A17B high · chat | 82.781.1 to 84.2 | 91.0 | 79.2 | 251.3 | 2.411 |
| Copernicus-31B | 81.880.1 to 83.4 | 93.1 | 77.0 | 0.184 | 0.134 |
| KnowLine-31B-Gen1.5 | 81.880.1 to 83.4 | 92.8 | 77.1 | 1.05 | No list price |
| KnowLine-31B-Gen2 | 81.279.5 to 82.8 | 93.4 | 76.0 | 0.168 | 0.120 |
| Turion Jev 31B | 80.178.4 to 81.7 | 92.7 | 74.8 | 0.405 | No list price |
| Inkling-Small high · chat · Jul 2026 | 80.078.3 to 81.6 | 87.2 | 76.9 | 97.7 | 1.137 |
| GLM-5.3-Flash reasoning on · chat | 78.877.0 to 80.5 | 89.3 | 74.3 | 102.6 | 0.543 |
| REAX 1.3 | 77.976.2 to 79.6 | 92.4 | 71.8 | 0.250 | 0.185 |
| Karpo Decision 31B | 77.876.0 to 79.5 | 93.4 | 71.1 | 0.159 | 0.105 |
| Gemini Decision 1 20261111 | 77.575.7 to 79.2 | 92.1 | 71.3 | 0.227 | 0.037 |
| d1-ultra | 77.175.3 to 78.8 | 92.1 | 70.7 | 0.223 | 0.014 |
| Logits Sigma-1 Duo | 75.073.2 to 76.8 | 91.9 | 67.8 | 0.250 | 0.181 |
| Qwen3.8-27B low · chat | 74.772.8 to 76.4 | 89.3 | 68.4 | 648.4 | 3.095 |
| GLM-5.3 low · chat | 74.572.7 to 76.3 | 87.7 | 68.9 | 17.1 | 0.814 |
| Sieve 31B v3 | 74.272.3 to 75.9 | 91.8 | 66.7 | 1.00 | No list price |
| Gemma 4 31B reasoning off · chat | 73.171.2 to 74.9 | 90.9 | 65.6 | 3.97 | 0.048 |
| Kimi K3.1 Pro no reasoning · chat | 72.770.8 to 74.5 | 92.5 | 64.3 | 5.75 | 0.853 |
| MiMo-V2.6-Flash high · chat | 72.170.2 to 73.9 | 86.6 | 66.0 | 21.4 | 2.219 |
| GLiDE 27B | 71.169.2 to 73.0 | 90.9 | 62.7 | 0.542 | No list price |
| Perplexity Decider 27B v1.2 API · with web | 68.766.8 to 70.6 | 88.9 | 60.2 | 0.482 | ≤ 1.034 |
| GLiDE v1.2 | 68.566.5 to 70.4 | 89.6 | 59.5 | 0.413 | 0.014 |
| Perplexity Decider 27B v1.3 | 68.466.5 to 70.3 | 89.0 | 59.7 | 0.110 | 0.081 |
| GPT-6 Luna low · chat | 68.366.3 to 70.2 | 87.3 | 60.2 | 7.71 | 0.048 |
| d1.2-ultra | 68.166.1 to 70.0 | 89.8 | 58.9 | 0.220 | 0.014 |
| GLiDE API | 67.965.9 to 69.8 | 87.8 | 59.4 | 2.09 | 0.160 |
| d3 | 67.865.8 to 69.7 | 89.2 | 58.7 | 0.092 | 0.070 |
| d3-ultra | 67.865.8 to 69.7 | 89.8 | 58.4 | 0.333 | 0.216 |
| Drex Max | 67.765.8 to 69.7 | 89.6 | 58.4 | 0.082 | 0.023 |
| Drex Max v1.1 | 67.465.4 to 69.3 | 88.6 | 58.4 | 0.313 | 0.023 |
| Mammoth-1 | 67.165.1 to 69.0 | 90.4 | 57.2 | 0.169 | 0.040 |
| Perplexity Decider 27B v1.2 DGX Spark | 67.165.1 to 69.0 | 89.2 | 57.7 | 0.580 | 0.008 |
| Perplexity Decider 27B v1.1 | 66.964.9 to 68.8 | 89.0 | 57.5 | 0.233 | 0.008 |
| Gemma 4 31B reasoning off · chat · 1 per request | 66.864.8 to 68.7 | 89.0 | 57.3 | 0.298 | 0.068 |
| MiniMax-M3.1-Flash high · chat | 66.564.5 to 68.4 | 82.9 | 59.5 | 112.6 | 0.814 |
| Perplexity Decider 27B v1.2 API | 66.364.3 to 68.3 | 89.0 | 56.7 | 0.254 | 0.008 |
| DeepSeek-V4.1-Flash reasoning off · chat · 1017 | 66.064.0 to 68.0 | 88.3 | 56.5 | 5.08 | 0.098 |
| DeepSeek-V4.1-Flash reasoning off · chat · first release | 65.963.9 to 67.9 | 87.8 | 56.6 | 4.36 | 0.098 |
| SubQ Snap Preview | 65.263.2 to 67.2 | 87.7 | 55.7 | 0.389 | 0.019 |
| SubQ Snap 1.1 | 63.961.8 to 65.9 | 87.0 | 54.0 | 0.378 | 0.015 |
| GPT-6 Luna Decisions API | 63.661.6 to 65.6 | 85.5 | 54.3 | 0.191 | 0.040 |
| Qwen3.8 Flash no reasoning · chat | 63.161.0 to 65.1 | 87.2 | 52.8 | 10.0 | 0.047 |
| KnowLine-12B-Gen1 | 62.560.4 to 64.5 | 86.0 | 52.5 | 0.071 | 0.052 |
| d1.2 | 61.759.7 to 63.7 | 86.0 | 51.4 | 0.213 | 0.007 |
| d1.1 | 61.559.5 to 63.5 | 84.6 | 51.7 | 0.195 | 0.007 |
| Claude Sonnet 5.5 low · chat | 61.459.3 to 63.4 | 82.9 | 52.2 | 3.87* | 1.896 |
| celeris-1-decision | 60.558.4 to 62.5 | 85.2 | 49.9 | 0.401 | 0.015 |
| Perplexity Decider 27B v1 | 60.458.3 to 62.4 | 86.0 | 49.5 | 0.216 | 0.016 |
| OneJev-27B | 59.657.5 to 61.6 | 85.2 | 48.7 | 0.125 | 0.099 |
| Jev 1.14 | 59.257.1 to 61.2 | 85.1 | 48.2 | 0.211 | 0.029 |
| Qwen3.5-4B reasoning on · chat | 58.956.8 to 61.0 | 79.1 | 50.3 | 140.3 | 2.998 |
| Mercury Decide | 58.356.2 to 60.3 | 84.5 | 47.1 | 0.313 | 0.008 |
| GLM-5.3-Flash reasoning off · chat | 57.855.7 to 59.8 | 82.2 | 47.4 | 1.07 | 0.047 |
| Gemma 4 E4B default reasoning · chat | 57.655.5 to 59.6 | 77.6 | 49.0 | 13.4 | 0.307 |
| Jev 1.13 | 57.555.4 to 59.6 | 83.7 | 46.4 | 0.243 | 0.029 |
| Clef | 57.355.2 to 59.3 | 85.1 | 45.5 | 0.205 | 0.110 |
| Claude Haiku 5.5 max · chat | 57.155.0 to 59.1 | 86.9 | 44.4 | 417.0 | 2.965 |
| Jeffan-9B-1211 | 57.054.9 to 59.1 | 82.9 | 46.0 | 0.064 | 0.044 |
| d1 | 57.054.9 to 59.1 | 82.8 | 46.0 | 0.305 | 0.007 |
| Drex 2.0 | 55.753.6 to 57.8 | 81.4 | 44.8 | 0.325 | 0.019 |
| Cygnet | 55.753.6 to 57.8 | 82.8 | 44.2 | 0.036 | 0.039 |
| Open-Jev-27B-v1.1 | 55.753.6 to 57.8 | 82.9 | 44.1 | 0.386 | 0.271 |
| Winnow-12B | 55.553.4 to 57.5 | 82.0 | 44.2 | 0.059 | 0.041 |
| KnowLine-9B-Gen1 | 54.552.5 to 56.6 | 79.9 | 43.8 | 0.057 | 0.041 |
| Cerno 9B v1.2 | 54.552.4 to 56.5 | 79.7 | 43.7 | 0.051 | 0.036 |
| Qwen3.8-27B vote share · reasoning off · chat | 54.352.2 to 56.4 | 84.0 | 41.7 | 1.62 | 0.825 |
| Qwen3.8-27B stated confidence · reasoning off · chat | 54.252.1 to 56.3 | 83.4 | 41.8 | 0.433 | 0.234 |
| KnowLine-9B-Gen2 | 53.351.2 to 55.4 | 79.6 | 42.1 | 0.048 | 0.035 |
| Adjudex 9B v5 | 53.151.0 to 55.2 | 80.8 | 41.4 | 0.038 | 0.026 |
| Elephant-1 | 52.450.3 to 54.4 | 80.5 | 40.4 | 0.074 | 0.014 |
| GLiDE 9B | 52.050.0 to 54.1 | 80.2 | 40.1 | 0.207 | No list price |
| Qwen3.8-27B yes/no scoring | 51.949.8 to 54.0 | 80.5 | 39.8 | 0.280 | 0.704 |
| Dexy-Decision-9B-V2 | 51.649.5 to 53.7 | 79.6 | 39.7 | 0.031 | 0.023 |
| StartLux-Decision-v1.1-9B-Instant | 51.249.1 to 53.3 | 78.8 | 39.5 | 0.154 | No list price |
| Perplexity Decider Flash v1.3 | 49.747.6 to 51.8 | 74.4 | 39.2 | 0.033 | 0.025 |
| vera-medium-omni | 49.147.0 to 51.2 | 75.5 | 37.9 | 0.029 | 0.021 |
| KnowLine 4B Gen5.5 | 48.546.4 to 50.6 | 74.8 | 37.3 | 0.099 | No list price |
| StartLux-Decision-v1.1-4B-Instant | 47.945.8 to 50.0 | 74.1 | 36.7 | 0.074 | No list price |
| Perplexity Decider Flash v1.2 | 47.845.7 to 49.9 | 72.7 | 37.2 | 0.132 | 0.004 |
| Valen-Decision-1.5-9B | 47.044.9 to 49.0 | 75.0 | 35.0 | 0.059 | 0.040 |
| Verdin-4B-v1.1 | 46.844.7 to 48.9 | 71.2 | 36.4 | 0.049 | 0.034 |
| Chinese Decision Pointer 4B | 46.744.6 to 48.8 | 70.9 | 36.4 | 0.035 | 0.025 |
| Microsoft-Decision-1 | 46.544.4 to 48.5 | 73.5 | 35.0 | 0.286 | 0.014 |
| Qwen3.5-4B reasoning off · chat | 46.244.1 to 48.3 | 68.4 | 36.7 | 32.2 | 0.894 |
| Clef-Flash | 45.743.6 to 47.8 | 73.0 | 34.1 | 0.075 | 0.017 |
| Qwen3-14B reasoning off · chat | 45.743.6 to 47.8 | 72.1 | 34.5 | 2.85 | 0.032 |
| Erk-14B reasoning off · chat | 45.042.9 to 47.1 | 72.3 | 33.4 | 2.39 | 0.106 |
| Metask-Jev-4B | 40.838.7 to 42.8 | 62.8 | 31.4 | 0.090 | 0.063 |
| KnowLine-2B-Gen1 | 36.834.8 to 38.8 | 54.3 | 29.4 | 0.062 | No list price |
| GLiDE 2B | 35.333.3 to 37.3 | 51.2 | 28.5 | 0.034 | No list price |
| StartLux-Decision-v1.1-2B-Instant | 34.332.3 to 36.3 | 52.9 | 26.3 | 0.040 | No list price |
| LFM2.5-2.6B default reasoning · chat | 33.931.9 to 35.9 | 45.4 | 29.0 | 51.2 | 2.761 |
| LFM3-1.7B reasoning on · chat | 32.430.5 to 34.4 | 44.2 | 27.4 | 24.2 | 1.277 |
| d1-2B | 29.427.5 to 31.3 | 40.9 | 24.5 | 0.019 | 0.013 |
| GLiDE 0.8B | 29.327.4 to 31.2 | 42.4 | 23.7 | 0.022 | No list price |
| OneJev1.6-0.8B | 29.127.3 to 31.1 | 39.6 | 24.6 | 0.017 | No list price |
| StartLux-Decision-v1.1-0.8B-Instant | 28.626.7 to 30.5 | 35.1 | 25.8 | 0.028 | No list price |
| LFM3-1.7B reasoning off · chat | 28.326.5 to 30.3 | 37.2 | 24.6 | 4.90 | 0.646 |
| Strands Decider 2B | 27.525.7 to 29.4 | 36.9 | 23.5 | 0.026 | 0.020 |
| KnowLine-0.8B-Gen1 | 27.125.3 to 29.0 | 38.3 | 22.4 | 0.033 | No list price |
| Decider 1 | 27.125.3 to 29.0 | 34.6 | 23.9 | 0.382 | 0.015 |
| Qwen3.5-2B reasoning off · chat | 25.423.6 to 27.2 | 34.3 | 21.6 | 8.84 | 0.268 |
| GLiClass v4 Edge | 23.321.6 to 25.2 | 25.9 | 22.2 | 0.025 | No list price |
| GLiDE 0.4B | 22.220.6 to 24.0 | 24.5 | 21.3 | 0.016 | No list price |
| LFM2.5-8B-A1B default reasoning · chat | 21.019.3 to 22.7 | 25.6 | 19.0 | 39.2 | 1.050 |
| GLiNER2.5-multi-Decide | 20.218.6 to 21.9 | 21.0 | 19.8 | 0.009 | 0.006 |
| Laya | 19.017.4 to 20.7 | 20.9 | 18.2 | 0.011 | 0.009 |
| Laya Multilingual | 17.916.3 to 19.5 | 17.1 | 18.2 | 0.009 | 0.008 |
| GLiNER2.5 Multi | 17.515.9 to 19.1 | 17.2 | 17.6 | 0.032 | 0.006 |
| MiniCPM5-2B reasoning off · chat | 5.64.7 to 6.6 | 6.6 | 5.2 | 172.0 | 5.464 |
Results 94aef0c1, updated 16 November 2026; answer key version 2.
The cost of every setup
Every setup, decision and chat models together · cheapest first, with accuracy, latency, the confidence cutoff and hallucination beside the cost · lower cost is better
Loading the table…
Accuracy against cost or latency
Up: accuracy, higher is better · across: cost or latency, lower is better
Loading charts…
2,198 questions · 115 models · 137 setups · updated 16 November 2026 · More about this page
Accuracy
The best chat setup gets 9.8 in 10 questions right; the best decision model 8.2 in 10
Share of the questions answered right · higher is better; the line is chance (20%)
Loading charts…
Every setup's accuracy, with tie groups: All setups · More about this finding
Non-hallucination
The decision model that chooses “I don't know” most does so on 95 in 100 of its answers that were not right; the chat model that does so most, on 76
Holding back by confidence
The answers each setup gave at 99% confidence or more, in its Text tab run, the bar their share of the 2,198 questions; how many of them were wrong; and its ranking score (AUROC: how often a right answer has higher confidence than a wrong one, 0.5 chance, 1 perfect). No wrong answer at 99% first, by answers kept, then by the share wrong among the kept answers; a setup with no answers at 99% last · more kept and fewer wrong is better
Decision models
Confidence: option probabilities.
Karpo Perplexity Gemini Mammoth vLLM-SR Fastino REAX Drex Elephant KnowLine Copernicus Liquid Logits Turion Jev Verdin StartLux Valen vera Cygnet OneJev Sieve SubQ Winnow Dexy Adjudex Chinese Decision Pointer Laya Cerno Clef GLiClass Jeffan Metask Open-Jev Strands
Chat models
Confidence: the confidence the model states for 38; option probabilities for 8; agreement across ten answers for 1.
Inkling GLM Qwen Claude Muse Gemini GPT DeepSeek Erk MiMo Kimi MiniMax Gemma Fastino Grok Confidence not measured (a plain answer line, no confidence): LFM3-1.7B · reasoning off · chat, LFM3-1.7B · reasoning on · chat, LFM2.5-8B-A1B · default reasoning · chat, Gemma 4 E4B · default reasoning · chat, MiniCPM5-2B · reasoning off · chat, LFM2.5-2.6B · default reasoning · chat, Qwen3.5-2B · reasoning off · chat, Qwen3.5-4B · reasoning off · chat and Qwen3.5-4B · reasoning on · chat.
Holding back with an “I don't know” option
“I don't know” among the answers that were not right, with it offered as a sixth option on the 2,198 questions; beside it the share answered (right + wrong of 2,198) and the error rate among the answers (wrong of right + wrong) · higher non-hallucination is better, read with the share answered; a model that always chose the option would reach 100% with no right answers
Decision models
Each of the 63 decision models' first run with it.
Chose “I don't know” on fewer than 1% of the 2,198 questions: Turion Jev 31B, 15 times; Gemini Decision 1 · 20261111, 10 times; StartLux-Decision-v1.1-4B-Instant, 21 times; Open-Jev-27B-v1.1, 14 times; d3-ultra, 10 times; Copernicus-31B, 5 times; Adjudex 9B v5, 8 times; Metask-Jev-4B, 10 times; Perplexity Decider 27B v1.3, 5 times; Karpo Decision 31B, 3 times; KnowLine-9B-Gen1, 5 times; Verdin-4B-v1.1, 5 times; REAX 1.3, 2 times; Perplexity Decider 27B v1.2 · API, 3 times; Logits Sigma-1 Duo, 2 times; Drex Max, 2 times; Perplexity Decider 27B v1.2 · DGX Spark, 2 times; KnowLine-31B-Gen1.5, 1 time; Cerno 9B v1.2, 2 times; KnowLine-12B-Gen1, 1 time; OneJev-27B, 1 time; Valen-Decision-1.5-9B, 1 time; KnowLine-31B-Gen2, 0 times; Sieve 31B v3, 0 times; d3, 0 times. Their rate here is low (3.3% or less) because they almost never chose the option; the part Holding back by confidence shows how their confidence holds back wrong answers.
Fastino Liquid Mammoth SubQ OneJev Drex Elephant Laya GLiClass StartLux Perplexity Chinese Decision Pointer Strands Clef vera KnowLine Jeffan Winnow Dexy Turion Jev Cygnet Gemini Open-Jev vLLM-SR Copernicus Adjudex Metask Karpo Verdin REAX Logits Cerno Valen Sieve
Chat models
Each of the 56 chat model setups' first run with it.
Liquid Grok Gemma MiniCPM Claude Qwen Gemini GPT Fastino MiniMax MiMo Inkling Muse Kimi DeepSeek GLM Erk Running: a chat model's runs with the option go English first, then Turkish, and each is added as it finishes; still running in Turkish: Claude Fable 5.2 · max · chat.
Holding back by confidence
The answers each setup gave at 99% confidence or more, in its Text tab run, the bar their share of the 2,198 questions; how many of them were wrong; and its ranking score (AUROC: how often a right answer has higher confidence than a wrong one, 0.5 chance, 1 perfect). No wrong answer at 99% first, by answers kept, then by the share wrong among the kept answers; a setup with no answers at 99% last · more kept and fewer wrong is better
Chat models
Confidence: the confidence the model states for 38; option probabilities for 8; agreement across ten answers for 1.
Inkling GLM Qwen Claude Muse Gemini GPT DeepSeek Erk MiMo Kimi MiniMax Gemma Fastino Grok Confidence not measured (a plain answer line, no confidence): LFM3-1.7B · reasoning off · chat, LFM3-1.7B · reasoning on · chat, LFM2.5-8B-A1B · default reasoning · chat, Gemma 4 E4B · default reasoning · chat, MiniCPM5-2B · reasoning off · chat, LFM2.5-2.6B · default reasoning · chat, Qwen3.5-2B · reasoning off · chat, Qwen3.5-4B · reasoning off · chat and Qwen3.5-4B · reasoning on · chat.
Holding back with an “I don't know” option
“I don't know” among the answers that were not right, with it offered as a sixth option on the 2,198 questions; beside it the share answered (right + wrong of 2,198) and the error rate among the answers (wrong of right + wrong) · higher non-hallucination is better, read with the share answered; a model that always chose the option would reach 100% with no right answers
Chat models
Each of the 56 chat model setups' first run with it.
Liquid Grok Gemma MiniCPM Claude Qwen Gemini GPT Fastino MiniMax MiMo Inkling Muse Kimi DeepSeek GLM Erk Running: a chat model's runs with the option go English first, then Turkish, and each is added as it finishes; still running in Turkish: Claude Fable 5.2 · max · chat.
Holding back by confidence
The answers each setup gave at 99% confidence or more, in its Text tab run, the bar their share of the 2,198 questions; how many of them were wrong; and its ranking score (AUROC: how often a right answer has higher confidence than a wrong one, 0.5 chance, 1 perfect). No wrong answer at 99% first, by answers kept, then by the share wrong among the kept answers; a setup with no answers at 99% last · more kept and fewer wrong is better
Decision models
Confidence: option probabilities.
Karpo Perplexity Gemini Mammoth vLLM-SR Fastino REAX Drex Elephant KnowLine Copernicus Liquid Logits Turion Jev Verdin StartLux Valen vera Cygnet OneJev Sieve SubQ Winnow Dexy Adjudex Chinese Decision Pointer Laya Cerno Clef GLiClass Jeffan Metask Open-Jev Strands
Holding back with an “I don't know” option
“I don't know” among the answers that were not right, with it offered as a sixth option on the 2,198 questions; beside it the share answered (right + wrong of 2,198) and the error rate among the answers (wrong of right + wrong) · higher non-hallucination is better, read with the share answered; a model that always chose the option would reach 100% with no right answers
Decision models
Each of the 63 decision models' first run with it.
Chose “I don't know” on fewer than 1% of the 2,198 questions: Turion Jev 31B, 15 times; Gemini Decision 1 · 20261111, 10 times; StartLux-Decision-v1.1-4B-Instant, 21 times; Open-Jev-27B-v1.1, 14 times; d3-ultra, 10 times; Copernicus-31B, 5 times; Adjudex 9B v5, 8 times; Metask-Jev-4B, 10 times; Perplexity Decider 27B v1.3, 5 times; Karpo Decision 31B, 3 times; KnowLine-9B-Gen1, 5 times; Verdin-4B-v1.1, 5 times; REAX 1.3, 2 times; Perplexity Decider 27B v1.2 · API, 3 times; Logits Sigma-1 Duo, 2 times; Drex Max, 2 times; Perplexity Decider 27B v1.2 · DGX Spark, 2 times; KnowLine-31B-Gen1.5, 1 time; Cerno 9B v1.2, 2 times; KnowLine-12B-Gen1, 1 time; OneJev-27B, 1 time; Valen-Decision-1.5-9B, 1 time; KnowLine-31B-Gen2, 0 times; Sieve 31B v3, 0 times; d3, 0 times. Their rate here is low (3.3% or less) because they almost never chose the option; the part Holding back by confidence shows how their confidence holds back wrong answers.
Fastino Liquid Mammoth SubQ OneJev Drex Elephant Laya GLiClass StartLux Perplexity Chinese Decision Pointer Strands Clef vera KnowLine Jeffan Winnow Dexy Turion Jev Cygnet Gemini Open-Jev vLLM-SR Copernicus Adjudex Metask Karpo Verdin REAX Logits Cerno Valen Sieve
Read non-hallucination with the share answered; the Hallucination tab has the full table. More about this finding
Cost
The median decision setup costs $0.025$24.63 per 1,000 questionsper 1M questions, the median chat setup $1.277$1,277
Estimated dollars per 1,000 questionsper 1M questions · lower is better
Loading charts…
The cost of every setup, with accuracy beside it · More about this finding
Latency
At 1 question per request, the median decision setup answers in 0.15 s, the median chat setup in 6.5 s
Median time per request · lower is better
Loading charts…
Reading vs grammar
122 of 137 setups score higher on reading than on grammar
Loading charts…
Calibration
d1.1 has the smallest calibration gap of the 77 decision models, as shipped: 2.0 points
How far a model’s confidence is from how often it is right, as shipped · lower is better
Loading charts…
Every setup's confidence: All setups, Confidence · More about this finding