AMVICC - MMVP Individual Accuracy: leaderboard
Vision-language evaluation from AMVICC (Basappa, Goel et al., 2026) on the 300 original MMVP questions about pairs of near-identical images, with free-form answers graded by GPT-4. The score is the unweighted mean of per-question accuracy (%) across MMVP's nine visual-pattern categories; higher is better.
Metric: Individual Accuracy (%). Source: arxiv.org. Saturation forecast: Not forecast. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.2 90B Vision Instruct | 91.88 | #1757 |
| 2 | Llama 4 Maverick | 87.39 | #1105 |
| 3 | GPT-4o | 85.06 | #741 |
| 4 | Gemini 2.5 Pro | 84.43 | #262 |
| 5 | Claude Opus 4.1 | 83.62 | #268 |
| 6 | Llama 4 Scout | 82.83 | #1787 |
| 7 | Qwen 2.5 VL 72B Instruct | 79.09 | #845 |
| 8 | pixtral-large-2411 | 78.41 | #1314 |
| 9 | Gemma 3 27B | 75.35 | #1638 |
| 10 | Claude Sonnet 4 | 74.91 | #397 |
| 11 | Grok 4 | 60.76 | #306 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=amvicc-mmvp-individual-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-12.