AA MMMU-Pro: leaderboard
Artificial Analysis's own run of MMMU-Pro, a multimodal exam built from MMMU by dropping questions a text-only model could answer and widening each question from four options to ten, covering six academic disciplines. Artificial Analysis asks 1,730 questions zero-shot in a single turn, extracts the answer letter by regular expression and reports pass@1 accuracy, higher is better. Source: arxiv.org/abs/2409.02813.
Metric: Accuracy (%). Source: artificialanalysis.ai. Saturation forecast: Around January 2027. 290 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 86.88 | #6 (GPT-6 Astra) |
| 2 | GPT-6 Astra (High) | 86.42 | #6 (GPT-6 Astra) |
| 3 | GPT-6 Astra (xHigh) | 86.24 | #6 (GPT-6 Astra) |
| 4 | GPT-6.1 Sol (Max) | 85.95 | #7 (GPT-6.1 Sol) |
| 5 | Gemini 3.8 Flash (High) | 85.61 | #21 (Gemini 3.8 Flash) |
| 6 | Gemini 3.7 Flash (High) | 85.49 | #23 (Gemini 3.7 Flash) |
| 7 | GPT-6 Astra (Medium) | 85.09 | #6 (GPT-6 Astra) |
| 8 | GPT-6.1 Sol (xHigh) | 85.09 | #7 (GPT-6.1 Sol) |
| 9 | GPT-6.1 Sol (High) | 84.91 | #7 (GPT-6.1 Sol) |
| 10 | Gemini 3.7 Flash (Low) | 84.86 | #23 (Gemini 3.7 Flash) |
| 11 | Claude Opus 5 (Max) | 84.74 | #12 (Claude Opus 5) |
| 12 | Gemini 3.7 Flash (Medium) | 84.74 | #23 (Gemini 3.7 Flash) |
| 13 | GPT-6 Astra (Low) | 84.62 | #6 (GPT-6 Astra) |
| 14 | Gemini 3.8 Flash (Low) | 84.51 | #21 (Gemini 3.8 Flash) |
| 15 | Gemini 3.5 Flash (High) | 84.28 | #60 (Gemini 3.5 Flash) |
No result here: #3 Claude Opus 5.5, #8 Gemini 4 Argon, #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aa-mmmu-pro · How It Works · Data refreshed daily, snapshot 2026-10-20.