MMLU-ProReasoning

An upgraded, harder MMLU with more than 12,000 questionsMMLU-Pro is an enhanced version of MMLU introduced in 2024 by the University of Toronto and other institutions. It increases answer choices from four to ten, filters questions susceptible to guessing, and adds reasoning-heavy items to raise difficulty and distinguish model performance more reliably.

Ranking Sources

#ModelScore

1

Gemini 3.1 ProGoogle

91.0%

2

Gemini 3 ProGoogle

90.1%

3

Claude Opus 4.6Anthropic

89.7%

4

Qwen3.7-MaxAlibaba

89.6%

5

Claude Opus 4.5Anthropic

89.5%

6

Qwen3.6-PlusAlibaba

88.5%

7

Qwen3.7-PlusAlibaba

88.5%

8

Qwen3.5-397B-A17BAlibaba

88.3%

9

Claude Sonnet 4.5Anthropic

88.2%

10

Kimi-K2.6Moonshot AI

88.1%

96 models total

Back to benchmarks