Decision model
Valen-Decision-1.5-9B
1 question per request
- Accuracy, all 2,198 questions
- 47.0%44.9% to 49.0%, 95% interval; 1,032 right
- Answers kept at 99% confidence
- 7 of 2,1980 wrong; share right 64.6% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.040$40.49Basis and price under Cost
- Latency
- 0.059 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Valen server (the card's valen-vllm plugin, Valen-Team/Valen at 773d3a6, in vllm/vllm-openai v0.33.0), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
valen- Parameters
- 9B dense, as the maker states it (counted from the released files: 8.40B).the whole checkpoint; source: counted from the checkpoint's safetensors, Valen-Team/Valen-Decision-1.5-9B@5a51211, huggingface.co/Valen-Team/Valen-Decision-1.5-9B; read 15 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- Valen-Team's released Valen-Decision-1.5-9B (Valen-Team/Valen-Decision-1.5-9B at 5a51211), a decision model on Qwen3.5-9B. It ran on our hardware through its card's vLLM deployment, the valen-vllm plugin installed without its dependencies into the official vLLM v0.33.0 image, which answers only to the model name 'valen', the key its runs carry; text only.
- Confidence from
- Option score
- Run
- 1275, 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
88e48d7a1055- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 47.0% | 44.9% to 49.0% | 1,032 of 2,198 |
| Reading | 75.0% | 71.5% to 78.2% | 492 of 656 |
| Grammar | 35.0% | 32.7% to 37.4% | 540 of 1,542 |
Comparison group: θ. 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
- 458 of 547, first English run (1278)
- Accuracy
- 83.7%
- Non-hallucination with “I don't know”
- 1.2%: 1 “I don't know” of 85 not right (run 1280)
- Answered with “I don't know”
- 99.8%: 546 of 547
- Error rate with “I don't know”
- 15.4%: 84 of 546 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.04140.55from GPU time
- Latency, median request, first English run
- 0.059 s; 1 request at once
- English run
- 1278
- 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
- 7, of which 0 wrong (share right: 64.6% to 100.0%, 95% interval); 2,191 skipped, the model less than 99% sure
- Calibration gap
- 3.8 points
- AUROC, all questions
- 0.77 0.75 to 0.79
- AUROC, reading
- 0.83 0.80 to 0.86
- AUROC, grammar
- 0.67 0.64 to 0.70
- Brier score
- 0.190
- Accuracy of its surest answers
- most confident fifth: 90.0% right; most confident half: 66.2% right; most confident four fifths: 52.9% right
- Typical score of a right answer
- 56.3%
- Typical score of a wrong answer
- 35.0%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.059 s (this run)
- Latency, 90th percentile
- 0.066 s
- Requests at once
- 1
- Throughput
- 1025.0 questions per minute of request time
- Request time
- 2.1 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
- $0.040$40.49
- 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
- 344
- Output tokens per question
- 0