Decision model
Cerno 9B v1.2
1 question per request
- Accuracy, all 2,198 questions
- 54.5%52.4% to 56.5%, 95% interval; 1,197 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
- $0.036$36.02Basis and price under Cost
- Latency
- 0.051 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Cerno server (a thin System One server around the card's CernoEngine, transformers: logic's for its English runs, compactor's di_serve.py for its Turkish runs), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
apertixai/cerno-9b-v1.2- Parameters
- 9B dense, as the maker states it (counted from the released files: 8.95B).language model with its LM head (text only); source: counted from the checkpoint's safetensors, apertixai/cerno-9b-v1.2@f02a632, huggingface.co/apertixai/cerno-9b-v1.2; read 10 November 2026
- Hardware
- 1 × RTX PRO 6000 Blackwell (96 GB); BF16
- Note
- Apertix AI's released Cerno 9B v1.2 (apertixai/cerno-9b-v1.2 at f02a632), a full fine-tune of Qwen3.5-9B with block-bidirectional attention. Its card ships no server, only an in-process engine (CernoEngine, transformers, BF16); it ran on our hardware, one request at a time, through a thin System One wrapper around that engine, text only: its English runs on one DGX Spark (logic's wrapper), its Turkish runs on one RTX PRO 6000 (compactor's di_serve.py, 15 November 2026). Input tokens are counted only where the engine reports them. Probabilities rounded to 4 decimals by its own code; they sum to 1 only within that rounding.
- Confidence from
- Option score
- Run
- 1201, 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 | 54.5% | 52.4% to 56.5% | 1,197 of 2,198 |
| Reading | 79.7% | 76.5% to 82.6% | 523 of 656 |
| Grammar | 43.7% | 41.3% to 46.2% | 674 of 1,542 |
Comparison group: VWXYZαβ. 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 (1085)
- Accuracy
- 85.9%
- Non-hallucination with “I don't know”
- 2.5%: 2 “I don't know” of 79 not right (run 1087)
- Answered with “I don't know”
- 99.6%: 545 of 547
- Error rate with “I don't know”
- 14.1%: 77 of 545 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- No list price
- Latency, median request, first English run
- 0.211 s; 1 request at once
- English run
- 1085
- Machine, English runs
- DGX Spark
- Date
- 13 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
- 9.8 points
- AUROC, all questions
- 0.78 0.76 to 0.80
- AUROC, reading
- 0.85 0.82 to 0.88
- AUROC, grammar
- 0.70 0.67 to 0.73
- Brier score
- 0.202
- Accuracy of its surest answers
- most confident fifth: 91.3% right; most confident half: 75.5% right; most confident four fifths: 61.4% right
- Typical score of a right answer
- 80.9%
- Typical score of a wrong answer
- 49.2%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.051 s (this run)
- Latency, 90th percentile
- 0.066 s
- Requests at once
- 1
- Throughput
- 1152.2 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
- $0.036$36.02
- Whole run
- $0.079
- 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
- 0
- Output tokens per question
- 0