Decision model
KnowLine-9B-Gen2
1 question per request
- Accuracy, all 2,198 questions
- 53.3%51.2% to 55.4%, 95% interval; 1,171 right
- Answers kept at 99% confidence
- No answers at this thresholdnone of the 2,198 questions at 99% or more
- Cost, $ per 1,000 questionsper 1M questions
- $0.035$35.40Basis and price under Cost
- Latency
- 0.048 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted
- Paid as
- Our own hardware
- Model ID
PelaAI/KnowLine-9B-Gen2- Parameters
- 9B dense, as the maker states it (counted from the released files: 9.65B).language model + the base's vision components (model-extra-from-base.safetensors; text only here); source: counted from the checkpoint's safetensors, PelaAI/KnowLine-9B-Gen2@8ec8eb6, huggingface.co/PelaAI/KnowLine-9B-Gen2; read 15 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- PelaAI's released checkpoint (13 November 2026), a fine-tune of Qwen3.5-9B, run on our hardware through its card's serve_knowline.sh (SGLang 0.5.23 in the lmsysorg/sglang image; the card's launcher names 0.5.21, then knowline_server.py), text only.
- Confidence from
- Option score
- Run
- 1306, 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
88e48d7a1055- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 53.3% | 51.2% to 55.4% | 1,171 of 2,198 |
| Reading | 79.6% | 76.3% to 82.5% | 522 of 656 |
| Grammar | 42.1% | 39.6% to 44.6% | 649 of 1,542 |
Comparison group: XYZαβγ. 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
- 466 of 547, first English run (1321)
- Accuracy
- 85.2%
- Non-hallucination with “I don't know”
- 1.3%: 1 “I don't know” of 79 not right (run 1323)
- Answered with “I don't know”
- 99.8%: 546 of 547
- Error rate with “I don't know”
- 14.3%: 78 of 546 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.03939.48from GPU time
- Latency, median request, first English run
- 0.056 s; 1 request at once
- English run
- 1321
- Machine, English runs
- RTX PRO 6000
- 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
- No answers at this threshold; 2,198 skipped, the model less than 99% sure
- Calibration gap
- 5.1 points
- AUROC, all questions
- 0.78 0.77 to 0.80
- AUROC, reading
- 0.83 0.80 to 0.87
- AUROC, grammar
- 0.70 0.67 to 0.73
- Brier score
- 0.186
- Accuracy of its surest answers
- most confident fifth: 92.9% right; most confident half: 74.9% right; most confident four fifths: 59.8% right
- Typical score of a right answer
- 66.1%
- Typical score of a wrong answer
- 38.6%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.048 s (this run)
- Latency, 90th percentile
- 0.066 s
- Requests at once
- 1
- Throughput
- 1172.5 questions per minute of request time
- Request time
- 1.9 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
- $0.035$35.40
- Whole run
- $0.078
- Basis
- Estimated from GPU time: the run's request time x the GPUs its server held x Runpod's on-demand price for an RTX PRO 6000, $2.49 per GPU-hour (read 5 November 2026). A cloud-equivalent estimate, not measured spending.
- Price
- $2.49 per GPU-hour, 1 GPU, for the run's request time (source, 2026-11-05)
- Input tokens per question
- 768
- Output tokens per question
- 1