Chat model
Claude Fable 5.2 max · chat
max reasoning · 16 questions per request
- Accuracy, all 2,198 questions
- 98.1%97.5% to 98.6%, 95% interval; 2,157 right
- Answers kept at 99% confidence
- 4 of 2,1980 wrong; share right 51.0% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $53.381$53,381Basis and price under Cost
- Latency
- 81.1 smedian request, 16 questions per request
How we called it
- Kind
- Chat model
- Access
- Claude Code CLI, Claude subscription
- Paid as
- Subscription
- Model ID
claude-fable-5-2- Parameters
- not disclosed by makerpages checked: www.anthropic.com/claude-fable-and-mythos-5-2, platform.claude.com/docs/en/models/overview; read 11 November 2026
- Confidence from
- Written estimate
- Run
- 980, 11 November 2026
Settings
- Reasoning
- max
- Request format
- 16 questions per request
- Requests at once
- 1
- Temperature
- not recorded
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
b55c22654bc7- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 98.1% | 97.5% to 98.6% | 2,157 of 2,198 |
| Reading | 97.7% | 96.3% to 98.6% | 641 of 656 |
| Grammar | 98.3% | 97.5% to 98.8% | 1,516 of 1,542 |
Comparison group: ab. 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
- 536 of 547, first English run (971)
- Accuracy
- 98.0%
- Non-hallucination with “I don't know”
- 28.6%: 4 “I don't know” of 14 not right (run 1032)
- Answered with “I don't know”
- 99.3%: 543 of 547
- Error rate with “I don't know”
- 1.8%: 10 of 543 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 52.94252,942subscription, priced as the API
- Latency, median request, first English run
- 74.0 s; 1 request at once
- English run
- 971
- Machine, English runs
- hosted API
- Date
- 11 November 2026
- Request format
- 16 questions per request
- Reasoning
- max
- Prompt version
5c3a4849752e- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 4, of which 0 wrong (share right: 51.0% to 100.0%, 95% interval); 2,194 skipped, the model less than 99% sure
- Calibration gap
- 7.5 points
- AUROC, all questions
- 0.90 0.84 to 0.94
- AUROC, reading
- 0.96 0.93 to 0.99
- AUROC, grammar
- 0.86 0.79 to 0.93
- Brier score
- 0.023
- Accuracy of its surest answers
- most confident fifth: 100.0% right; most confident half: 99.8% right; most confident four fifths: 99.6% right
- Typical score of a right answer
- 93.0%
- Typical score of a wrong answer
- 70.0%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 81.1 s (this run)
- Latency, 90th percentile
- 393.0 s
- Requests at once
- 1
- Throughput
- 5.6 questions per minute of request time
- Request time
- 391.5 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
- $53.381$53,381
- Whole run
- $117.33
- Basis
- Subscription run priced as if we had used the API
- Price
- $8 input, $0.20 cached input, $10 5-minute cache writes, $16 1-hour cache writes, $40 output per 1M tokens (source, 2026-11-11)
- Input tokens per question
- 564
- Output tokens per question
- 1,165