MMMU Pro
Multimodal · 2024-09-04
MMMU-Pro is a more robust version of MMMU that removes text-only-answerable items, expands the candidate options, and adds a vision-only input setting. Scores are reported as answer accuracy.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 84.7 |
| Gemini 3.5 Flash | 84.3 |
| Gemini 3.6 Flash | 83.2 |
| GPT-5.6 Sol | 82.7 |
| Claude Fable 5 | 81.2 |
| GPT-5.5 | 81.2 |
| GPT-5.6 Terra | 80.7 |
| Muse Spark | 80.5 |
| Qwen3.7-Plus | 80.5 |
| Grok 4.5 | 80.4 |
| Gemini 3 Pro Preview | 80.2 |
| Gemini 3 Flash Preview | 79.9 |
| Kimi K2.6 | 79.4 |
| Gemini 3.5 Flash-Lite | 79.0 |
| Opus 4.8 | 78.9 |
| Opus 4.7 | 78.8 |
| GPT-5.6 Luna | 78.6 |
| MiniMax M3 | 78.6 |
| GPT-5.3-Codex | 78.5 |
| Kimi K3 | 78.5 |
| Grok 4.3 | 78.1 |
| GPT-5.4 | 78.0 |
| Qwen3.6 Plus (2026-04-02) | 78.0 |
| Sonnet 5 | 77.3 |
| Qwen3.5 397B A17B | 77.3 |
| Grok Build 0.1 | 76.5 |
| GPT-5.2-Codex | 76.3 |
| Gemini 3.1 Flash-Lite Preview | 75.5 |
| GPT-5.1 | 75.5 |
| MiMo-V2.5 | 75.4 |
| Kimi K2.5 | 75.4 |
| Step 3.7 Flash | 75.3 |
| Qwen3.5-27B | 75.0 |
| Qwen3.6 35B A3B | 75.0 |
| Qwen3.5 122B A10B | 75.0 |
| Gemini 2.5 Pro | 74.9 |
| Qwen3.6 27B | 74.6 |
| GPT-5.2 | 74.6 |
| Grok 4.20 | 74.6 |
| Sonnet 4.6 | 74.5 |