MMLU-PRO mmlu-pro Leaderboard
Successor to MMLU - 12K questions, harder distractors, less contamination. · Metric: Accuracy (higher is better) · 🟢 Updated 18h ago
| # | Model | Accuracy | Paper |
|---|---|---|---|
| 1 | Gemini 3 Pro Preview (high) | 89.80 | — |
| 2 | Claude Opus 4.5 (Reasoning) | 89.50 | — |
| 3 | Gemini 3 Pro Preview (low) | 89.50 | — |
| 4 | Gemini 3 Flash Preview (Reasoning) | 89.00 | — |
| 5 | Claude Opus 4.5 (Non-reasoning) | 88.90 | — |
| 6 | Gemini 3 Flash Preview (Non-reasoning) | 88.20 | — |
| 7 | Claude 4.1 Opus (Reasoning) | 88.00 | — |
| 8 | Claude 4.5 Sonnet (Reasoning) | 87.50 | — |
| 9 | MiniMax-M2.1 | 87.50 | — |
| 10 | GPT-5.2 (xhigh) | 87.40 | — |
| 11 | Claude 4 Opus (Reasoning) | 87.30 | — |
| 12 | GPT-5 (high) | 87.10 | — |
| 13 | GPT-5.1 (high) | 87.00 | — |
| 14 | GPT-5 (medium) | 86.70 | — |
| 15 | Grok 4 | 86.60 | — |
| 16 | GPT-5 Codex (high) | 86.50 | — |
| 17 | DeepSeek V3.2 Speciale | 86.30 | — |
| 18 | DeepSeek V3.2 (Reasoning) | 86.20 | — |
| 19 | Gemini 2.5 Pro | 86.20 | — |
| 20 | Claude 4.5 Sonnet (Non-reasoning) | 86.00 | — |
| 21 | Claude 4 Opus (Non-reasoning) | 86.00 | — |
| 22 | GPT-5.1 Codex (high) | 86.00 | — |
| 23 | GPT-5 (low) | 86.00 | — |
| 24 | GPT-5.2 (medium) | 85.90 | — |
| 25 | Gemini 2.5 Pro Preview (Mar' 25) | 85.80 | — |
| 26 | GLM-4.7 (Reasoning) | 85.60 | — |
| 27 | Doubao Seed Code | 85.40 | — |
| 28 | Grok 4.1 Fast (Reasoning) | 85.40 | — |
| 29 | o3 | 85.30 | — |
| 30 | DeepSeek V3.1 (Reasoning) | 85.10 | — |
| 31 | Extending Item Response Theory for Efficient and Meaningful Multilingual Evaluation | 11.00 | link |