Image JevBench v0.3.0 · individual system
Standard One 8B
adapter · 8BBase model: mistralai/Ministral-3-8B-Instruct-2512-BF16sourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
Not measured
Listed, not ranked.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 27.7
- calibration
- 75.4
- speed
- 83.3
- cost
- Not measured
Cost and speed evidence
- Cost per 1,000 decisions
- Not measured
- Cost basis
- Cost basis not reported.
- Measured latency · p50 / p95
- 0.210 s / 0.331 s
- Adjusted latency · p50 / p95
- 0.571 s / 0.812 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 50.6%
- Sealed accuracy
- 51.5%
- Measurement note
- measured-unrankable: CORE cost unavailable (runtime returns no GPU-seconds), so no composite
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.