Image JevBench v0.1.5 · individual system
JevAny-27B SFT
adapter · 27BBase model: Qwen/Qwen3.8-27Bsource 1source 2Archived v0.1.5 score, not ranked on the v0.3 scale. See the v0.1.5 archive.
local/GPU evaluation
Image JevBench v0.1.5 composite score
63.795
v0.1.5 rank #18 of 50 (archived).
From the published full-benchmark aggregate. The composite combines intelligence, calibration, speed and cost.
Published axes
- intelligence
- 69.6
- calibration
- 90.0
- speed
- 86.1
- cost
- 48.1
Cost and speed evidence
- Cost per 1,000 decisions
- USD 0.053910
- Cost basis
- measured GPU seconds x $1.19/GPU-hour
- Measured latency · p50 / p95
- 0.161 s / 0.185 s
- Adjusted latency · p50 / p95
- 0.471 s / 0.519 s
- Latency adjustment
- v1.4 local latency adjustment
- Public accuracy
- 77.6%
- Sealed accuracy
- 80.9%
Published 2026-10-01. Scores and ranks can change in a later release. See the method notes.