Image JevBench v0.3.0 · individual system
OmniJev 4B (tinnel123)
adapter · 4BBase model: Qwen/Qwen3.5-4Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
14.831
Rank #18 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 28.0
- calibration
- 66.4
- speed
- 86.4
- cost
- 49.3
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.048856
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.146 s / 0.186 s
- Adjusted latency · p50 / p95
- 0.442 s / 0.522 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 52.8%
- Sealed accuracy
- 50.8%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.