AA MMLU-Pro — leaderboard
Artificial Analysis independent evaluation of MMLU-Pro: 12,000 graduate-level questions across 14 subjects with 10 answer options. 280+ models tested.
Metric: Accuracy (%). Source: artificialanalysis.ai. Status: saturated. 345 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 89.76 |
| 2 | Claude Opus 4.5 (Thinking) | 89.45 |
| 3 | Claude Opus 4.5 (Non-reasoning) | 88.92 |
| 4 | Gemini 3 Flash (Preview) (Non-reasoning) | 88.22 |
| 5 | Claude Opus 4.1 (Thinking) | 87.99 |
| 6 | MiniMax-M2.1 | 87.49 |
| 7 | Claude Sonnet 4.5 (Thinking) | 87.45 |
| 8 | GPT-5.2 (xHigh) | 87.36 |
| 9 | Claude Opus 4 (Thinking) | 87.32 |
| 10 | GPT-5 (High) | 87.07 |
| 11 | GPT-5.1 (High) | 87.04 |
| 12 | GPT-5 (Medium) | 86.74 |
| 13 | Grok 4 | 86.58 |
| 14 | GPT-5 Codex (High) | 86.49 |
| 15 | DeepSeek V3.2 Speciale | 86.31 |
Interactive version: theaggregate.ai/benchmark?slug=aa-mmlu-pro · How the rankings work · Data refreshed daily, snapshot 2026-07-22.