Image JevBench v0.3.0 · individual system
Gevva E2B multimodal
2BBase model: Gemma 4 E2Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
2.019
Rank #36 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 12.8
- calibration
- 81.9
- speed
- 85.1
- cost
- 48.8
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.050794
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.160 s / 0.258 s
- Adjusted latency · p50 / p95
- 0.470 s / 0.665 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 41.7%
- Sealed accuracy
- 40.9%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.