MMMU-Pro Extended (Mercor held-out)
Multimodal
Mercor's held-out Extended twin of MMMU-Pro: 140 Mercor multimodal questions that require joint visual and textual reasoning, scored by answer accuracy. The public twin is MMMU Pro (mmmu-pro). Mercor does not publish the held-out tasks or say how they were drawn or written, so this is its own benchmark, not a component of the public one.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6.1 Sol | 61.3 |
| GPT-6 Astra | 60.0 |
| GPT-5.6 Sol Pro | 51.9 |
| GPT-6 Sol | 51.9 |
| Claude Opus 5.5 | 51.0 |
| Claude Fable 5.1 | 47.6 |
| GPT-5.6 Terra | 47.4 |
| GPT-5.5 | 43.8 |
| GPT-5.4 | 39.8 |
| GPT-5.6 Luna | 39.2 |
| Claude Fable 5 | 37.6 |
| Gemini 3.7 Flash | 36.4 |
| Gemini 3.8 Flash | 36.4 |
| Kimi K3 | 36.0 |
| Muse Spark 1.1 | 35.5 |
| GPT-6 Luna | 34.8 |
| Claude Opus 5 | 34.3 |
| Muse Spark 1.3 | 33.8 |
| Claude Haiku 5.5 | 31.9 |
| Opus 4.8 | 31.4 |
| Gemini 3.1 Pro Preview | 31.0 |
| Muse Spark 1.2 | 30.2 |
| Grok 4.5 | 29.8 |
| Gemini 3.6 Flash | 29.0 |
| Gemini 3.5 Flash | 28.8 |
| Opus 4.7 | 28.3 |
| Grok 4.6 | 26.0 |
| DeepSeek V4.1 Flash | 25.0 |
| Qwen3.5 397B A17B | 24.3 |
| Inkling | 24.0 |
| Kimi K2.7 Code | 23.1 |
| GLM-5.3 Flash | 22.6 |
| Sonnet 5 | 20.7 |
| Sonnet 4.6 | 19.9 |
| MiniMax M3 | 18.1 |