Image JevBench v0.3.0 · individual system

djev-dev BF16

26B-A4BBase model: google/diffusiongemma-26B-A4B-itsource

Published source

local/GPU evaluation

Image JevBench v0.3.0 composite score

0.039

Rank #40 of 44 ranked systems.

From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.

Published axes

intelligence
3.9
calibration
46.9
speed
83.1
cost
35.4

Cost and speed evidence

Cost per 1,000 decisions
USD 0.071545
Cost basis
measured GPU seconds x $1.19/GPU-hour
Measured latency · p50 / p95
0.220 s / 0.344 s
Adjusted latency · p50 / p95
0.591 s / 0.838 s
Latency adjustment
v1.4 local latency adjustment
Public accuracy
33.6%
Sealed accuracy
36.1%

Published 2026-10-06. Scores and ranks can change in a later release. See the method notes.