Image JevBench v0.3.0 · individual system
OpenJev 4B NLI v2 (official image-premise path)
4BBase model: Qwen/Qwen3.5-4Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
0.000
Rank #43 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 16.4
- calibration
- 0.0
- speed
- 75.1
- cost
- 33.9
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.155263
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.554 s / 1.157 s
- Adjusted latency · p50 / p95
- 1.259 s / 2.463 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 43.4%
- Sealed accuracy
- 43.8%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.