Chat model
Qwen3.8 Flash no reasoning · chat
reasoning none · 16 questions per request
- Accuracy, all 2,198 questions
- 63.1%61.0% to 65.1%, 95% interval; 1,386 right
- Answers kept at 99% confidence
- 55 of 2,1989 wrong; share right 71.7% to 91.1%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.047$47.13Basis and price under Cost
- Latency
- 10.0 smedian request, 16 questions per request
How we called it
- Kind
- Chat model
- Access
- OpenRouter chat API, Alibaba's endpoint, fallbacks off, reasoning off
- Paid as
- Provider service
- Model ID
qwen/qwen3.8-flash- Parameters
- not disclosed by makerpages checked: www.alibabacloud.com/help/en/model-studio/model-pricing, openrouter.ai/qwen/qwen3.8-flash; read 11 November 2026
- Note
- Alibaba's closed chat model (API id qwen3.8-flash), added on the reading bench as on backgammon and Logic & Math
- Confidence from
- Written estimate
- Run
- 985, 11 November 2026
Settings
- Reasoning
- none
- Request format
- 16 questions per request
- Requests at once
- 4
- Temperature
- not recorded
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
3b4daaeaddf3- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 63.1% | 61.0% to 65.1% | 1,386 of 2,198 |
| Reading | 87.2% | 84.4% to 89.5% | 572 of 656 |
| Grammar | 52.8% | 50.3% to 55.3% | 814 of 1,542 |
Comparison group: IJKLMNOP. 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
- 476 of 547, first English run (986)
- Accuracy
- 87.0%
- Non-hallucination with “I don't know”
- 5.6%: 4 “I don't know” of 71 not right (run 988)
- Answered with “I don't know”
- 99.3%: 543 of 547
- Error rate with “I don't know”
- 12.3%: 67 of 543 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.05049.69at list prices
- Latency, median request, first English run
- 10.1 s; 4 requests at once
- English run
- 986
- Machine, English runs
- hosted API
- Date
- 11 November 2026
- Request format
- 16 questions per request
- Reasoning
- none
- Prompt version
5c3a4849752e- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 55, of which 9 wrong (share right: 71.7% to 91.1%, 95% interval); 2,143 skipped, the model less than 99% sure
- Calibration gap
- 28.7 points
- AUROC, all questions
- 0.66 0.64 to 0.68
- AUROC, reading
- 0.68 0.62 to 0.74
- AUROC, grammar
- 0.64 0.62 to 0.67
- Brier score
- 0.305
- Accuracy of its surest answers
- most confident fifth: 79.4% right; most confident half: 75.2% right; most confident four fifths: 68.3% right
- Typical score of a right answer
- 95.0%
- Typical score of a wrong answer
- 90.0%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 10.0 s (this run)
- Latency, 90th percentile
- 12.1 s
- Requests at once
- 4
- Throughput
- 384.6 questions per minute of request time
- Request time
- 5.7 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
- $0.047$47.13
- Whole run
- $0.10
- Basis
- Estimated from published prices and recorded token usage
- Price
- $0.15 input, $0.016 cached input, $0.47 output per 1M tokens (source, 2026-11-11)
- Input tokens per question
- 237
- Output tokens per question
- 25