Decision model
KnowLine-9B-Gen1
1 question per request
- Accuracy, all 2,198 questions
- 54.5%52.5% to 56.6%, 95% interval; 1,199 right
- Answers kept at 99% confidence
- 62 of 2,1980 wrong; share right 94.2% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.041$40.69Basis and price under Cost
- Latency
- 0.057 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-Gen1- 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-Gen1@325378d, huggingface.co/PelaAI/KnowLine-9B-Gen1; read 15 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- PelaAI's released checkpoint, a merged LoRA 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
- 1174, 15 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
815031979442- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 54.5% | 52.5% to 56.6% | 1,199 of 2,198 |
| Reading | 79.9% | 76.6% to 82.8% | 524 of 656 |
| Grammar | 43.8% | 41.3% to 46.3% | 675 of 1,542 |
Comparison group: VWXYZα. 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
- 470 of 547, first English run (1177)
- Accuracy
- 85.9%
- Non-hallucination with “I don't know”
- 1.3%: 1 “I don't know” of 77 not right (run 1179)
- Answered with “I don't know”
- 99.8%: 546 of 547
- Error rate with “I don't know”
- 13.9%: 76 of 546 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.04241.55from GPU time
- Latency, median request, first English run
- 0.059 s; 1 request at once
- English run
- 1177
- Machine, English runs
- RTX PRO 6000
- Date
- 15 November 2026
- Request format
- 1 question per request
- Reasoning
- off
- Prompt version
47f5776e71db- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 62, of which 0 wrong (share right: 94.2% to 100.0%, 95% interval); 2,136 skipped, the model less than 99% sure
- Calibration gap
- 4.6 points
- AUROC, all questions
- 0.78 0.77 to 0.80
- AUROC, reading
- 0.85 0.82 to 0.88
- AUROC, grammar
- 0.70 0.68 to 0.73
- Brier score
- 0.187
- Accuracy of its surest answers
- most confident fifth: 95.7% right; most confident half: 74.8% right; most confident four fifths: 61.5% right
- Typical score of a right answer
- 73.0%
- Typical score of a wrong answer
- 41.8%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.057 s (this run)
- Latency, 90th percentile
- 0.078 s
- Requests at once
- 1
- Throughput
- 1019.9 questions per minute of request time
- Request time
- 2.2 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
- $0.041$40.69
- Whole run
- $0.089
- 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