HELM MMLU - Astronomy: leaderboard

Metric: EM. Source: crfm.stanford.edu. 79 models tracked.

Top models

#ModelScore
1Claude 3.5 Sonnet (20241022)97.37
2Claude 3 Opus (20240229)96.71
3Claude 3.5 Sonnet (20240620)96.05
4GPT-4o (2024-08-06)94.74
5GPT-4 Turbo94.08
6GPT-4o (2024-05-13)94.08
7GPT-4 Turbo (Preview)94.08
8Gemini 1.5 Pro (002)93.42
9Qwen 2.5 72B Instruct93.42
10Qwen 2 72B Instruct93.42
11GPT-4 (0613)93.42
12Gemini 2.0 Flash (Preview)92.76
13Llama 3.2 90B Vision Instruct92.11
14DeepSeek V392.11
15Llama 3.1 405B Instruct92.11

Interactive version: theaggregate.ai/benchmark?slug=helm-mmlu-astronomy · How It Works · Data refreshed daily, snapshot 2026-09-08.