Image JevBench v0.3.0 · individual system
OmniJev-Qwen3.5-9B-v4 (tzcfly)
adapter · 9BBase model: Qwen/Qwen3.5-9Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
14.988
Rank #17 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 29.0
- calibration
- 37.3
- speed
- 89.5
- cost
- 57.6
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.025876
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.074 s / 0.115 s
- Adjusted latency · p50 / p95
- 0.298 s / 0.379 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 54.0%
- Sealed accuracy
- 51.1%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.