Image JevBench v0.3.0 · individual system
OmniJev 2B (tinnel123)
adapter · 2BBase model: Qwen/Qwen3.5-2Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
3.168
Rank #34 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 15.0
- calibration
- 60.9
- speed
- 87.6
- cost
- 52.3
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.039000
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.116 s / 0.155 s
- Adjusted latency · p50 / p95
- 0.381 s / 0.459 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 41.3%
- Sealed accuracy
- 43.5%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.