Image JevBench v0.3.0 · individual system
Jevify Qwen3-VL-2B T2
adapter · 2BBase model: Qwen/Qwen3-VL-2B-Instructsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
3.287
Rank #33 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 15.1
- calibration
- 72.8
- speed
- 87.0
- cost
- 53.5
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.035569
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.094 s / 0.223 s
- Adjusted latency · p50 / p95
- 0.338 s / 0.597 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 43.0%
- Sealed accuracy
- 42.6%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.