Image JevBench v0.3.0 · individual system
Visual-Jev generic (zero-shot Qwen3.5-4B baseline)
4BBase model: undisclosedlocal/GPU evaluation
Image JevBench v0.3.0 composite score
11.136
Rank #22 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 29.7
- calibration
- 83.8
- speed
- 80.6
- cost
- 40.4
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.091354
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.266 s / 0.570 s
- Adjusted latency · p50 / p95
- 0.683 s / 1.290 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 53.2%
- Sealed accuracy
- 52.3%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.