Image JevBench v0.3.0 · individual system
JPT-0.8B (kirp / llm2jev)
adapter · 0.8BBase model: Qwen3.5-0.8Bsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
7.171
Rank #27 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 20.4
- calibration
- 71.7
- speed
- 87.1
- cost
- 53.7
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.034996
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.095 s / 0.208 s
- Adjusted latency · p50 / p95
- 0.341 s / 0.567 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 46.4%
- Sealed accuracy
- 46.3%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.