Decision model
OneJev1.6-0.8B
1 question per request
- Accuracy, all 2,198 questions
- 29.1%27.3% to 31.1%, 95% interval; 640 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
- No list priceBasis and price under Cost
- Latency
- 0.017 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
OmniJev/OneJev1.6-0.8B- Parameters
- 800M dense, as the maker states it (counted from the released files: 0.85B).language model (tied embeddings) + 0.10B vision encoder; source: counted from the checkpoint's safetensors, OmniJev/OneJev1.6-0.8B@bcd88a3, huggingface.co/OmniJev/OneJev1.6-0.8B; read 11 November 2026
- Hardware
- 1 × DGX Spark, GB10 (128 GB unified memory); BF16
- Note
- OmniJev's released OneJev1.6-0.8B (OmniJev/OneJev1.6-0.8B at bcd88a3, whose files but its README are those of c89933c), an average of three full fine-tunes of Qwen3.5-0.8B that also reads images. It ran on our hardware (one DGX Spark), one request at a time, through the server its card names (Flide 0.1.11, flide serve), with vision on (torchvision 0.29.1). Probabilities rounded to 6 decimals by Flide; they sum to 1 only within that rounding.
- Confidence from
- Option score
- Run
- 1029, 11 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
b49f0758f6a5- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 29.1% | 27.3% to 31.1% | 640 of 2,198 |
| Reading | 39.6% | 36.0% to 43.4% | 260 of 656 |
| Grammar | 24.6% | 22.6% to 26.9% | 380 of 1,542 |
Comparison group: σφψ. Compare with another setup.
Accuracy by topic
Vision (preliminary)
Preliminary: these numbers are not final and may change. This setup is condition A of the Vision tab, which asks the 763 questions with underlined or numbered words as text (A), as a page image (B), as both (C) and through OCR models (D); the share answered right, its 95% interval in the number's tooltip.
| Model | A · text onlyaccuracy, higher is better | B · image onlyaccuracy, higher is better | C · text plus imageaccuracy, higher is better | Best OCR routeaccuracy, higher is better |
|---|---|---|---|---|
| OneJev1.6-0.8B Open · 0.8B | 28.0% | 26.5% | 28.4% | 27.4%via Surya OCR 2 |
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
- 325 of 547, first English run (1025)
- Accuracy
- 59.4%
- Non-hallucination with “I don't know”
- 28.4%: 74 “I don't know” of 261 not right (run 1027)
- Answered with “I don't know”
- 86.5%: 473 of 547
- Error rate with “I don't know”
- 39.5%: 187 of 473 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- No list price
- Latency, median request, first English run
- 0.018 s; 1 request at once
- English run
- 1025
- Machine, English runs
- DGX Spark
- Date
- 11 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
- 9.3 points
- AUROC, all questions
- 0.61 0.58 to 0.63
- AUROC, reading
- 0.62 0.57 to 0.66
- AUROC, grammar
- 0.56 0.53 to 0.59
- Brier score
- 0.210
- Accuracy of its surest answers
- most confident fifth: 45.0% right; most confident half: 34.2% right; most confident four fifths: 31.3% right
- Typical score of a right answer
- 36.6%
- Typical score of a wrong answer
- 32.6%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.017 s (this run)
- Latency, 90th percentile
- 0.022 s
- Requests at once
- 1
- Throughput
- 3363.7 questions per minute of request time
- Request time
- 0.7 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
- No list price
- Basis
- No dollar estimate; see hardware and time
- Input tokens per question
- 396
- Output tokens per question
- 1
Going first in front of an LLM
Replay of recorded answers, cutoff chosen on half the questions, averaged over the splits. Savings are estimates from API prices; the Method page explains how.
| LLM behind it | Share taken | Accuracy change | Saving |
|---|---|---|---|
| Gemini 3.8 Flash high · chat | 0.4% | −0.13 points | 0%† |
| GPT-6 Astra high · chat | 0.4% | −0.13 points | 0%† |
| GPT-6.1 Sol high · chat | 0.5% | −0.11 points | 1%† |
| GPT-6 Astra low · chat | 0.4% | −0.13 points | 0%† |
| GPT-6 Astra low · chat · 1 per request | 0.2% | −0.10 points | 0%† |
| Gemini 3.8 Flash low · chat · 1 per request | 0.2% | −0.10 points | 0%† |
| GPT-6.1 Sol low · chat · 1 per request | 0.4% | −0.13 points | 0%† |
| GPT-6.1 Sol low · chat | 0.5% | −0.11 points | 1%† |
| Claude Opus 5.5 high · chat | 0.4% | −0.13 points | 0%† |
| GPT-6.1 Terra high · chat | 0.2% | −0.10 points | 0%† |
| Claude Opus 5.5 low · chat · 1 per request | 0.2% | −0.10 points | 0%† |
| Gemini 3.8 Flash low · chat | 0.2% | −0.10 points | 0%† |
| GPT-6.1 Terra low · chat | 0.2% | −0.10 points | 0%† |
| Claude Opus 5.5 low · chat | 0.4% | −0.13 points | 0%† |
| Claude Sonnet 5.5 high · chat · 1 per request | 0.4% | −0.13 points | 0%† |
| Claude Sonnet 5.5 high · chat | 0.5% | −0.11 points | 1%† |
† Leaves out the first model’s hardware cost, which we have not estimated.