Decision model
Copernicus-31B
1 question per request
- Accuracy, all 2,198 questions
- 81.8%80.1% to 83.4%, 95% interval; 1,798 right
- Answers kept at 99% confidence
- 107 of 2,1980 wrong; share right 96.5% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.134$134Basis and price under Cost
- Latency
- 0.184 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Copernicus server (the card's copernicus_server.py, Transformers), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
Copernicus-31B- Parameters
- 31B dense, as the maker states it (counted from the released files: 31.27B).the whole checkpoint, its vision encoder included (text only here); the card says approximately 31 billion; source: counted from the checkpoint's safetensors, Poindexter-Labs/Copernicus-31B@be71525, huggingface.co/Poindexter-Labs/Copernicus-31B; read 15 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- Poindexter Labs' released checkpoint, a merge of KnowLine-31B-Gen1 and Sieve-31B-v3 (both on Gemma-4-31B-it), run on our hardware through its card's copernicus_server.py (Transformers, text only, the card's defaults).
- Confidence from
- Option score
- Run
- 1165, 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 | 81.8% | 80.1% to 83.4% | 1,798 of 2,198 |
| Reading | 93.1% | 90.9% to 94.8% | 611 of 656 |
| Grammar | 77.0% | 74.8% to 79.0% | 1,187 of 1,542 |
Comparison group: nop. 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
- 508 of 547, first English run (1168)
- Accuracy
- 92.9%
- Non-hallucination with “I don't know”
- 0.0%: 0 “I don't know” of 39 not right (run 1170)
- Answered with “I don't know”
- 100.0%: 547 of 547
- Error rate with “I don't know”
- 7.1%: 39 of 547 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.142142from GPU time
- Latency, median request, first English run
- 0.188 s; 1 request at once
- English run
- 1168
- 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
- 107, of which 0 wrong (share right: 96.5% to 100.0%, 95% interval); 2,091 skipped, the model less than 99% sure
- Calibration gap
- 9.1 points
- AUROC, all questions
- 0.88 0.86 to 0.89
- AUROC, reading
- 0.94 0.91 to 0.96
- AUROC, grammar
- 0.84 0.82 to 0.86
- Brier score
- 0.115
- Accuracy of its surest answers
- most confident fifth: 100.0% right; most confident half: 98.3% right; most confident four fifths: 91.1% right
- Typical score of a right answer
- 85.0%
- Typical score of a wrong answer
- 44.4%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.184 s (this run)
- Latency, 90th percentile
- 0.253 s
- Requests at once
- 1
- Throughput
- 308.8 questions per minute of request time
- Request time
- 7.1 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
- $0.134$134
- Whole run
- $0.30
- 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
- 737
- Output tokens per question
- 1