Image JevBench v0.3.0 · individual system
Qevi-2B (MeerDevelopment)
adapter · 2BBase model: Qwen/Qwen3-VL-2B-Instructsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
0.025
Rank #41 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 2.7
- calibration
- 19.6
- speed
- 86.4
- cost
- 52.3
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.038900
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.099 s / 0.257 s
- Adjusted latency · p50 / p95
- 0.348 s / 0.664 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 32.8%
- Sealed accuracy
- 35.3%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.