Image JevBench v0.3.0 · individual system
Jev-Vision 8B (SeanLiu)
adapter · 8BBase model: Qwen/Qwen3-VL-8B-Instructsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
25.508
Rank #9 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 39.0
- calibration
- 36.5
- speed
- 84.8
- cost
- 47.4
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.056483
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.191 s / 0.234 s
- Adjusted latency · p50 / p95
- 0.531 s / 0.618 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 60.4%
- Sealed accuracy
- 58.0%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.