AdaptMMBench (Oracle Visual Evidence): leaderboard
Metric: Overall accuracy (%) on the 1,300 AdaptMMBench samples (auxiliary-line problems evaluated separately), enhancement-transformed samples included, in oracle-visual mode, where the annotated gold visual evidence (the zoomed or restored region) is supplied and the model reasons in text; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 90.08 | #264 |
| 2 | Gemini 3 Pro | 89.85 | #77 |
| 3 | GPT-5 | 88.69 | #91 |
| 4 | Qwen 3 VL 32B Instruct | 85.62 | #276 |
| 5 | Qwen 3 VL 8B Instruct | 76.85 | #401 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=adaptmmbench-oracle-visual-evidence · How It Works · Data refreshed daily, snapshot 2026-10-11.