Decision model
Gemini Decision 1 20261111
1 question per request
- Accuracy, all 2,198 questions
- 77.5%75.7% to 79.2%, 95% interval; 1,704 right
- Answers kept at 99% confidence
- 318 of 2,1980 wrong; share right 98.8% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.037$37.19Basis and price under Cost
- Latency
- 0.227 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- OpenRouter decisions API, Google AI Studio endpoint
- Paid as
- Provider service
- Model ID
google/gemini-decision-1- Variant
- via OpenRouter
- Parameters
- not disclosed by makerpages checked: aistudio.google.com/docs/models/gemini-decision-1, openrouter.ai/google/gemini-decision-1, blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/; read 11 November 2026
- Note
- Google's decision model, called through OpenRouter's System One route on its one endpoint, Google AI Studio. It returns a probability for each option and bills input tokens only.
- Confidence from
- Option score
- Run
- 1021, 11 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
b49f0758f6a5- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 77.5% | 75.7% to 79.2% | 1,704 of 2,198 |
| Reading | 92.1% | 89.8% to 93.9% | 604 of 656 |
| Grammar | 71.3% | 69.0% to 73.5% | 1,100 of 1,542 |
Comparison group: qrstuv. 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 |
|---|---|---|---|---|
| Gemini Decision 1 20261111 Closed · Efficient | 76.1% | 66.4% | 78.4% | 70.9%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
- 498 of 547, first English run (1009)
- Accuracy
- 91.0%
- Non-hallucination with “I don't know”
- 9.1%: 5 “I don't know” of 55 not right (run 1019)
- Answered with “I don't know”
- 99.1%: 542 of 547
- Error rate with “I don't know”
- 9.2%: 50 of 542 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.04039.51at list prices
- Latency, median request, first English run
- 0.235 s; 1 request at once
- English run
- 1009
- Machine, English runs
- hosted API
- Date
- 11 November 2026
- Request format
- 1 question per request
- Reasoning
- off
- Prompt version
47f5776e71db- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 318, of which 0 wrong (share right: 98.8% to 100.0%, 95% interval); 1,880 skipped, the model less than 99% sure
- Calibration gap
- 2.0 points
- AUROC, all questions
- 0.86 0.84 to 0.88
- AUROC, reading
- 0.90 0.86 to 0.94
- AUROC, grammar
- 0.83 0.81 to 0.85
- Brier score
- 0.122
- Accuracy of its surest answers
- most confident fifth: 100.0% right; most confident half: 97.2% right; most confident four fifths: 86.4% right
- Typical score of a right answer
- 93.8%
- Typical score of a wrong answer
- 52.3%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.227 s (this run)
- Latency, 90th percentile
- 0.354 s
- Requests at once
- 1
- Throughput
- 197.5 questions per minute of request time
- Request time
- 11.1 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
- $0.037$37.19
- Whole run
- $0.082
- Basis
- Estimated from published prices and recorded token usage
- Price
- $0.10 input, output free per 1M tokens (source, 2026-11-11)
- Input tokens per question
- 372
- 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 | 23.6% | −0.09 points | 22% |
| GPT-6 Astra high · chat | 27.9% | −0.17 points | 27% |
| GPT-6.1 Sol high · chat | 33.6% | −0.17 points | 31% |
| GPT-6 Astra low · chat | 31.0% | −0.17 points | 30% |
| GPT-6 Astra low · chat · 1 per request | 31.4% | −0.16 points | 31% |
| Gemini 3.8 Flash low · chat · 1 per request | 29.0% | −0.15 points | 25% |
| GPT-6.1 Sol low · chat · 1 per request | 26.5% | −0.15 points | 26% |
| GPT-6.1 Sol low · chat | 34.4% | −0.18 points | 31% |
| Claude Opus 5.5 high · chat | 41.1% | −0.13 points | 41% |
| GPT-6.1 Terra high · chat | 35.5% | −0.18 points | 28% |
| Claude Opus 5.5 low · chat · 1 per request | 47.9% | −0.16 points | 48% |
| Gemini 3.8 Flash low · chat | 31.2% | −0.16 points | 17% |
| GPT-6.1 Terra low · chat | 33.9% | −0.20 points | 25% |
| Claude Opus 5.5 low · chat | 54.1% | −0.19 points | 53% |
| Claude Sonnet 5.5 high · chat · 1 per request | 62.2% | −0.13 points | 62% |
| Claude Sonnet 5.5 high · chat | 77.9% | −0.03 points | 77% |