Image JevBench v0.3.0 · individual system
diffusiongemma-26b djev v10 step160 (snowicarus)
fine-tune · 26B-A4BBase model: google/diffusiongemma-26B-A4B-itsourcelocal/GPU evaluation
Image JevBench v0.3.0 composite score
5.903
Rank #29 of 44 ranked systems.
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 23.2
- calibration
- 50.3
- speed
- 82.9
- cost
- 41.2
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.071013
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.224 s / 0.354 s
- Adjusted latency · p50 / p95
- 0.599 s / 0.857 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 47.2%
- Sealed accuracy
- 48.8%
Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.