MMErroR - Error Type Classification - Mathematics & Logic: leaderboard
MMErroR (Shi et al., ACL 2026) error type classification on its Mathematics & Logic samples: naming which of four error types (visual perception, knowledge deployment, question comprehension, reasoning) the single injected error in an image-grounded reasoning chain belongs to. Accuracy (%); higher is better.
Metric: Error Type Classification Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 66.37 | #111 |
| 2 | Seed 2.0 Pro | 65.47 | #209 |
| 3 | GPT-5.2 (xHigh) | 64.56 | #172 (GPT-5.2) |
| 4 | Kimi K2.5 | 63.66 | #252 |
| 5 | Claude Opus 4.5 | 62.76 | #152 |
| 6 | Grok 4 | 61.56 | #306 |
| 7 | Qwen 3 VL 32B (Thinking) | 59.46 | #557 |
| 8 | Qwen-VL-Max | 53.45 | #1226 |
| 9 | GLM-4.6V | 51.05 | #698 |
| 10 | Qwen 3 VL 32B Instruct | 47.45 | #582 |
| 11 | Llama 4 Scout Instruct | 43.84 | #1532 |
| 12 | Llama 4 Maverick Instruct | 42.64 | #1125 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mmerror-error-type-classification-mathematics-logic · How It Works · Data refreshed daily, snapshot 2026-11-12.