MMMU-Pro (Vals 4-Option)
Multimodal
Vals AI evaluates approximately 1,700 questions from the MMMU-Pro dataset using a four-option multiple-choice format, the official chain-of-thought prompt, and pass@1 scoring. This source-specific protocol is distinct from canonical ten-option MMMU-Pro and from standard MMMU.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 90.6 |
| Claude Opus 5 | 89.9 |
| Claude Fable 5 | 89.3 |
| Gemini 3.8 Flash | 89.1 |
| Gemini 3.7 Flash | 89.0 |
| GPT-5.6 Sol | 88.8 |
| Gemini 3.6 Flash | 88.4 |
| Gemini 3.5 Flash | 88.3 |
| GPT-5.5 | 88.3 |
| Gemini 3.1 Pro Preview | 88.2 |
| Kimi K3 | 88.2 |
| Qwen3.8-Max | 88.0 |
| Gemini 3 Flash Preview | 87.6 |
| Gemini 3 Pro Preview | 87.5 |
| GPT-5.4 | 87.5 |
| Muse Spark | 87.4 |
| GPT-5.2 | 86.7 |
| Opus 4.8 | 86.6 |
| Muse Spark 1.1 | 86.6 |
| GPT-5.6 Terra | 86.5 |
| Kimi K2.6 | 86.3 |
| Muse Spark 1.2 | 86.1 |
| GLM-5.3 Flash | 86.0 |
| Opus 4.7 | 85.5 |
| GPT-5.6 Luna | 85.0 |
| Kimi K2.5 | 84.3 |
| Qwen3.6 Plus (2026-04-02) | 84.2 |
| Qwen3.8 27B | 83.9 |
| Opus 4.6 | 83.9 |
| Sonnet 4.6 | 83.6 |
| Gemini 3.5 Flash-Lite | 83.6 |
| Grok 4.20 | 83.5 |
| GPT-5.1 | 83.2 |
| Grok 4.3 | 83.1 |
| Sonnet 5 | 83.0 |
| Opus 4.5 | 82.9 |
| Gemini 3.1 Flash-Lite Preview | 82.5 |
| Qwen3.5-Flash (source unspecified) | 81.9 |
| GPT-5 | 81.5 |
| Gemini 2.5 Pro Experimental 03-25 | 81.3 |