AMVICC - MMVP Pair Accuracy - Structural and Physical Characteristics: leaderboard
AMVICC (Basappa, Goel et al., 2026) asks vision-language models the original MMVP questions about pairs of nearly identical images, grading free-form answers with GPT-4 against the authors' rubric. This board is pair accuracy in percent on the 15 image pairs of MMVP's Structural and Physical Characteristics category, where a pair counts only when both of its questions are answered correctly; higher is better.
Metric: Pair accuracy (%; the 15 MMVP image pairs in the Structural and Physical Characteristics category, a pair credited only when both of its questions are answered correctly; free-form answers graded by GPT-4 against the authors' rubric). Source: arxiv.org. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.2 90B Vision Instruct | 93.33 | #1749 |
| 2 | Llama 4 Maverick | 86.67 | #1099 |
| 3 | Llama 4 Scout | 86.67 | #1776 |
| 4 | Qwen 2.5 VL 72B Instruct | 73.33 | #846 |
| 5 | pixtral-large-2411 | 73.33 | #1309 |
| 6 | GPT-4o | 66.67 | #741 |
| 7 | Gemini 2.5 Pro | 60 | #266 |
| 8 | Claude Sonnet 4 | 60 | #400 |
| 9 | Claude Opus 4.1 | 60 | #272 |
| 10 | Grok 4 | 53.33 | #307 |
| 11 | Gemma 3 27B | 46.67 | #1631 |
Interactive version: theaggregate.ai/benchmark?slug=amvicc-mmvp-pair-accuracy-structural-and-physical-characteristics · How It Works · Data refreshed daily, snapshot 2026-11-13.