Image JevBench v0.3.0 · individual system
Jev-Omni
fine-tune · 12BBase model: Gemma 4 12B ITsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
32.845
Rank #4 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 37.5
- calibration
- 71.0
- speed
- 88.8
- cost
- 60.0
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.021544
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.087 s / 0.129 s
- Adjusted latency · p50 / p95
- 0.325 s / 0.409 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 60.0%
- Sealed accuracy
- 56.8%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.