Decision model
d1.2-ultra
1 question per request
- Accuracy, all 2,198 questions
- 68.1%66.1% to 70.0%, 95% interval; 1,497 right
- Answers kept at 99% confidence
- 11 of 2,1980 wrong; share right 74.1% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.014$14.34Basis and price under Cost
- Latency
- 0.220 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Liquid AI decisions API
- Paid as
- Provider service
- Model ID
d1.2-ultra- Parameters
- not disclosed by makerpages checked: www.liquid.ai/blog/d1-2; read 16 November 2026
- Confidence from
- Option score
- Run
- 1347, 16 November 2026
Settings
- Reasoning
- off
- Request format
- 1 question per request
- Requests at once
- 1
- Temperature
- not recorded
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
0f7123a34772- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 68.1% | 66.1% to 70.0% | 1,497 of 2,198 |
| Reading | 89.8% | 87.2% to 91.9% | 589 of 656 |
| Grammar | 58.9% | 56.4% to 61.3% | 908 of 1,542 |
Comparison group: ABCDEG. Compare with another setup.
Accuracy by topic
Vision (preliminary)
Preliminary: these numbers are not final and may change. This setup is condition A of the Vision tab, which asks the 763 questions with underlined or numbered words as text (A), as a page image (B), as both (C) and through OCR models (D); the share answered right, its 95% interval in the number's tooltip.
| Model | A · text onlyaccuracy, higher is better | B · image onlyaccuracy, higher is better | C · text plus imageaccuracy, higher is better | Best OCR routeaccuracy, higher is better |
|---|---|---|---|---|
| d1.2-ultra Closed · Efficient | 64.1% | 45.3% | 66.7% | 61.7%via Surya OCR 2 |
English
The Text tab's 547 kept reading questions translated into English, asked with this setup's Text tab settings. Every setup: English reading; against Turkish: English and Turkish.
- Right answers
- 494 of 547, first English run (1359)
- Accuracy
- 90.3%
- Non-hallucination with “I don't know”
- 89.3%: 109 “I don't know” of 122 not right (run 1363)
- Answered with “I don't know”
- 80.1%: 438 of 547
- Error rate with “I don't know”
- 3.0%: 13 of 438 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.01413.98at list prices
- Latency, median request, first English run
- 0.221 s; 1 request at once
- English run
- 1359
- Machine, English runs
- hosted API
- Date
- 16 November 2026
- Request format
- 1 question per request
- Reasoning
- off
- Prompt version
47f5776e71db- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 11, of which 0 wrong (share right: 74.1% to 100.0%, 95% interval); 2,187 skipped, the model less than 99% sure
- Calibration gap
- 3.4 points
- AUROC, all questions
- 0.82 0.81 to 0.84
- AUROC, reading
- 0.94 0.92 to 0.96
- AUROC, grammar
- 0.75 0.72 to 0.77
- Brier score
- 0.157
- Accuracy of its surest answers
- most confident fifth: 99.8% right; most confident half: 91.0% right; most confident four fifths: 75.4% right
- Typical score of a right answer
- 81.7%
- Typical score of a wrong answer
- 43.3%
Repeats
Run 5 times on all questions.
- Answers changed in 5 runs
- 1 of 2,198
- Same wrong answer in all 5
- 700 of 2,198
- Wrong in at least one run
- 701 of 2,198
- Accuracy in each run
- 68.1%, 68.1%, 68.1%, 68.1%, 68.1%
Latency
- Latency, median request
- 0.220 s (this run)
- Latency, 90th percentile
- 0.321 s
- Requests at once
- 1
- Throughput
- 236.4 questions per minute of request time
- Request time
- 9.3 minutes, over the requests whose answers were kept
Latency: the time from sending one request to receiving its answer, measured from our computer, network and any retries within the request included. Request time: the time at least one request was in flight, over the requests whose answers were kept. It is not the run's elapsed time: it leaves out requests that failed and any pause before a run was resumed.
Cost
- Cost per 1,000 questionsper 1M questions
- $0.014$14.34
- Whole run
- $0.032
- Basis
- Estimated from published prices and recorded token usage
- Price
- $0.042 input, output free per 1M tokens (source, 2026-11-16)
- Input tokens per question
- 341
- Output tokens per question
- 0
Going first in front of an LLM
Replay of recorded answers, cutoff chosen on half the questions, averaged over the splits. Savings are estimates from API prices; the Method page explains how.
| LLM behind it | Share taken | Accuracy change | Saving |
|---|---|---|---|
| Gemini 3.8 Flash high · chat | 24.4% | −0.13 points | 24% |
| GPT-6 Astra high · chat | 23.6% | −0.12 points | 23% |
| GPT-6.1 Sol high · chat | 25.2% | −0.16 points | 24% |
| GPT-6 Astra low · chat | 23.6% | −0.12 points | 23% |
| GPT-6 Astra low · chat · 1 per request | 22.8% | −0.10 points | 23% |
| Gemini 3.8 Flash low · chat · 1 per request | 28.1% | −0.19 points | 27% |
| GPT-6.1 Sol low · chat · 1 per request | 27.9% | −0.15 points | 28% |
| GPT-6.1 Sol low · chat | 23.6% | −0.12 points | 22% |
| Claude Opus 5.5 high · chat | 22.9% | −0.11 points | 23% |
| GPT-6.1 Terra high · chat | 23.8% | −0.14 points | 21% |
| Claude Opus 5.5 low · chat · 1 per request | 27.8% | −0.18 points | 28% |
| Gemini 3.8 Flash low · chat | 27.2% | −0.23 points | 22% |
| GPT-6.1 Terra low · chat | 33.5% | −0.16 points | 30% |
| Claude Opus 5.5 low · chat | 31.2% | −0.26 points | 31% |
| Claude Sonnet 5.5 high · chat · 1 per request | 31.5% | −0.19 points | 31% |
| Claude Sonnet 5.5 high · chat | 48.9% | −0.16 points | 49% |