Chat model
GPT-6 Astra medium · chat
medium reasoning · 16 questions per request
- Accuracy, all 2,198 questions
- 97.5%96.8% to 98.1%, 95% interval; 2,143 right
- Answers kept at 99% confidence
- 1,749 of 2,1988 wrong; share right 99.1% to 99.8%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $5.447$5,447Basis and price under Cost
- Latency
- 15.3 smedian request, 16 questions per request
How we called it
- Kind
- Chat model
- Access
- Codex CLI, ChatGPT subscription
- Paid as
- Subscription
- Model ID
gpt-6-astra- Parameters
- not disclosed by makerpages checked: developers.openai.com/api/docs/models/gpt-6-astra; read 9 November 2026
- Confidence from
- Written estimate
- Run
- 610, 5 November 2026
Settings
- Reasoning
- medium
- Request format
- 16 questions per request
- Requests at once
- 1
- Temperature
- not recorded
Reasoning effort levels (OpenAI's model page, read 28 October 2026): low, medium, high, xhigh, max. Top level: max. This run: medium.
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
d4268ef9b103- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 97.5% | 96.8% to 98.1% | 2,143 of 2,198 |
| Reading | 97.0% | 95.3% to 98.0% | 636 of 656 |
| Grammar | 97.7% | 96.9% to 98.4% | 1,507 of 1,542 |
Comparison group: abcde. Compare with another setup.
Accuracy by topic
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
- 525 of 547, first English run (624)
- Accuracy
- 96.0%
- Non-hallucination with “I don't know”
- 16.7%: 4 “I don't know” of 24 not right (run 813)
- Answered with “I don't know”
- 99.3%: 543 of 547
- Error rate with “I don't know”
- 3.7%: 20 of 543 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 4.6964,696subscription, priced as the API
- Latency, median request, first English run
- 10.7 s; 1 request at once
- English run
- 624
- Machine, English runs
- hosted API
- Date
- 5 November 2026
- Request format
- 16 questions per request
- Reasoning
- medium
- Prompt version
5c3a4849752e- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 1,749, of which 8 wrong (share right: 99.1% to 99.8%, 95% interval); 449 skipped, the model less than 99% sure
- Calibration gap
- 1.0 points
- AUROC, all questions
- 0.90 0.85 to 0.94
- AUROC, reading
- 0.97 0.95 to 0.98
- AUROC, grammar
- 0.86 0.80 to 0.91
- Brier score
- 0.021
- Accuracy of its surest answers
- most confident fifth: 100.0% right; most confident half: 99.7% right; most confident four fifths: 99.5% right
- Typical score of a right answer
- 99.0%
- Typical score of a wrong answer
- 94.0%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 15.3 s (this run)
- Latency, 90th percentile
- 30.6 s
- Requests at once
- 1
- Throughput
- 49.5 questions per minute of request time
- Request time
- 44.4 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
- $5.447$5,447
- Whole run
- $11.97
- Basis
- Subscription run priced as if we had used the API
- Price
- $10 input, $1 cached input, $12.50 cache writes, $50 output per 1M tokens (source, 2026-10-08)
- Input tokens per question
- 529
- Output tokens per question
- 23