JevK5 Release - JevBench hard (public half): leaderboard
JevK5 Release - JevBench hard (public half): Accuracy fraction (0-1). JevK5 official card: distinct trained v0.2 and v0.3 checkpoints (v0.2 retained under its Hub tag), Qwen3.5-4B decision model. JevBench easy/original/hard split accuracy, native external teacher held-out validation. Training-split MASSIVE/BANKING evaluation is withheld, not presented as held-out capability. Inference runtime updates are not separate models.
Metric: Accuracy fraction (0-1). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | JevK5 v0.3 | 0.78 | |
| 2 | JevK5 v0.2 | 0.74 |
Interactive version: theaggregate.ai/benchmark?slug=jevk5-release-jevbench-hard-public-half · How It Works · Data refreshed daily, snapshot 2026-11-12.