MMLU: leaderboard
Massive Multitask Language Understanding: 15,908 questions across 57 subjects from STEM to humanities. The original broad knowledge benchmark for LLMs, now considered saturated at the frontier.
Metric: Accuracy (%). Source: epoch.ai. Saturation forecast: Estimated already saturated. 136 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-11-20) | 88.1 | #381 |
| 2 | Claude 3.5 Sonnet (20241022) | 87.3 | #291 |
| 3 | DeepSeek V3 | 87.2 | #321 |
| 4 | Gemini 1.5 Pro (002) | 86.9 | #349 |
| 5 | Claude 3.5 Sonnet (20240620) | 86.5 | #299 |
| 6 | GPT-4 (0314) | 86.4 | #574 |
| 7 | Llama 3.3 70B Instruct | 86.3 | #534 |
| 8 | Gemini 1.5 Pro (001) | 85.9 | #444 |
| 9 | Qwen 2.5 72B Instruct | 85.3 | #454 |
| 10 | Qwen 2.5 72B | 85 | #499 |
| 11 | Phi-4 | 84.8 | #706 |
| 12 | Claude 3 Opus (20240229) | 84.6 | #412 |
| 13 | Llama 3.1 405B Instruct | 84.5 | #457 |
| 14 | Llama 3.1 405B | 84.4 | #466 |
| 15 | GPT-4o (2024-08-06) | 84.3 | #336 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmlu · How It Works · Data refreshed daily, snapshot 2026-10-20.