HELM MMLU - Prehistory: leaderboard

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

Top models

#ModelScore
1Claude 3.5 Sonnet (20240620)95.06
2Claude 3.5 Sonnet (20241022)94.14
3Claude 3 Opus (20240229)94.14
4GPT-4o (2024-05-13)93.83
5GPT-4o (2024-08-06)93.52
6Llama 3.1 405B Instruct92.9
7Gemini 1.5 Pro (002)92.59
8Nova Pro92.59
9GPT-4 (0613)92.59
10DeepSeek V392.28
11GPT-4 Turbo91.98
12Mistral Large 2 (Jul)91.98
13GPT-4 Turbo (Preview)91.67
14Qwen 2 72B Instruct91.36
15Qwen 2.5 72B Instruct91.05

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