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. Status: saturated. 136 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Human Expert | 90 |
| 2 | GPT-4o (2024-11-20) | 88.1 |
| 3 | Claude 3.5 Sonnet (20241022) | 87.3 |
| 4 | DeepSeek V3 | 87.2 |
| 5 | Gemini 1.5 Pro (002) | 86.9 |
| 6 | Claude 3.5 Sonnet (20240620) | 86.5 |
| 7 | GPT-4 (0314) | 86.4 |
| 8 | Llama 3.3 70B Instruct | 86.3 |
| 9 | Gemini 1.5 Pro (001) | 85.9 |
| 10 | Qwen 2.5 72B Instruct | 85.3 |
| 11 | Qwen 2.5 72B | 85 |
| 12 | Phi-4 | 84.8 |
| 13 | Claude 3 Opus (20240229) | 84.6 |
| 14 | Llama 3.1 405B Instruct | 84.5 |
| 15 | Llama 3.1 405B | 84.4 |
Interactive version: theaggregate.ai/benchmark?slug=mmlu · How the rankings work · Data refreshed daily, snapshot 2026-07-22.