AutoTrust JEV VL Release - ERQA Spatial reasoning: leaderboard
AutoTrust JEV VL Release - ERQA Spatial reasoning: ERQA accuracy (%). AutoTrust JEV-27B-VL official card: System 1 is the trained decision artifact, System 2 uses unmodified Qwen3.8-27B reasoning fused 50/50 with System 1 probabilities; low/medium hybrid runs remain explicit configurations of the released JEV product, never plain Qwen observations. Owner-run ERQA 400-question categories, MicroLens zero-shot recommendation, and vision reward/judge evaluation are retained with native protocols. Source-copied competitor columns and ambiguous joint JEV-27B/JEV-27B-VL text aggregates are withheld. Benchmark-native published comparator rows do not establish this release evaluated those other models.
Metric: ERQA accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | JEV-27B-VL | 53.6 | #219 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=autotrust-jev-vl-release-erqa-spatial-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.