HELM MMLU - Astronomy: leaderboard
Metric: EM. Source: crfm.stanford.edu. 79 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 97.37 |
| 2 | Claude 3 Opus (20240229) | 96.71 |
| 3 | Claude 3.5 Sonnet (20240620) | 96.05 |
| 4 | GPT-4o (2024-08-06) | 94.74 |
| 5 | GPT-4 Turbo | 94.08 |
| 6 | GPT-4o (2024-05-13) | 94.08 |
| 7 | GPT-4 Turbo (Preview) | 94.08 |
| 8 | Gemini 1.5 Pro (002) | 93.42 |
| 9 | Qwen 2.5 72B Instruct | 93.42 |
| 10 | Qwen 2 72B Instruct | 93.42 |
| 11 | GPT-4 (0613) | 93.42 |
| 12 | Gemini 2.0 Flash (Preview) | 92.76 |
| 13 | Llama 3.2 90B Vision Instruct | 92.11 |
| 14 | DeepSeek V3 | 92.11 |
| 15 | Llama 3.1 405B Instruct | 92.11 |
Interactive version: theaggregate.ai/benchmark?slug=helm-mmlu-astronomy · How It Works · Data refreshed daily, snapshot 2026-09-08.