Decision model
Jeffan-9B-1211
1 question per request
- Accuracy, all 2,198 questions
- 57.0%54.9% to 59.1%, 95% interval; 1,253 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
- $0.044$44.29Basis and price under Cost
- Latency
- 0.064 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
Jeffan-9B-1211- Parameters
- 9B dense, as the maker states it (counted from the released files: 8.95B).language model, text only; the card says bf16, 8.95B parameters; source: counted from the checkpoint's safetensors, tattrongvu/Jeffan-9B-1211@862d1a7, huggingface.co/tattrongvu/Jeffan-9B-1211; read 15 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- tattrongvu's released Jeffan-9B-1211 (tattrongvu/Jeffan-9B-1211 at 862d1a7), built on Qwen3.5-9B, which reads every option's probability from one forward pass and generates no text. Its card ships only a Decision Index engine (jeffan_engine:DecisionEngine), so it ran on our hardware behind compactor's di_serve.py (transformers 5.17.0, flash-linear-attention 0.5.2, without causal-conv1d, so PyTorch's reference convolution), text only.
- Confidence from
- Option score
- Run
- 1192, 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
a75966fe551d- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 57.0% | 54.9% to 59.1% | 1,253 of 2,198 |
| Reading | 82.9% | 79.9% to 85.6% | 544 of 656 |
| Grammar | 46.0% | 43.5% to 48.5% | 709 of 1,542 |
Comparison group: STUVWXY. 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
- 479 of 547, first English run (1195)
- Accuracy
- 87.6%
- Non-hallucination with “I don't know”
- 0.0%: 0 “I don't know” of 71 not right (run 1197)
- Answered with “I don't know”
- 100.0%: 547 of 547
- Error rate with “I don't know”
- 13.0%: 71 of 547 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.04645.85from GPU time
- Latency, median request, first English run
- 0.067 s; 1 request at once
- English run
- 1195
- 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
- No answers at this threshold; 2,198 skipped, the model less than 99% sure
- Calibration gap
- 4.6 points
- AUROC, all questions
- 0.80 0.78 to 0.82
- AUROC, reading
- 0.87 0.84 to 0.90
- AUROC, grammar
- 0.72 0.69 to 0.74
- Brier score
- 0.182
- Accuracy of its surest answers
- most confident fifth: 98.2% right; most confident half: 79.3% right; most confident four fifths: 64.2% right
- Typical score of a right answer
- 64.5%
- Typical score of a wrong answer
- 37.5%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.064 s (this run)
- Latency, 90th percentile
- 0.073 s
- Requests at once
- 1
- Throughput
- 937.0 questions per minute of request time
- Request time
- 2.3 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
- $0.044$44.29
- Whole run
- $0.097
- 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
- 393
- Output tokens per question
- 0