Pass IndexThe State of AISign in

MMLU-Pro

MMLU-Pro is run by TIGER-Lab. It has ranked 262 models, of which the catalogue holds 21, scoring from 86.4 to 91.16 on Overall (accuracy).

measured by TIGER-Lab · the board itself

PlaceModelMetricScore
1stof 262Gemini 3.1 Pro PreviewOverall (accuracy)91.16
2ndof 262gemini-3-pro-previewOverall (accuracy) (11/25)90.1
3rdof 262o1Overall (accuracy)89.3
4thof 262Claude Opus 4.6Overall (accuracy) (thinking)89.1
5thof 262Gemini 3 Flash PreviewOverall (accuracy) (12/25)88.6
6thof 262MiniMax M2.1Overall (accuracy)88
7thof 262Qwen3.5 397B A17BOverall (accuracy)87.8
8thof 262Seed-2.0-LiteOverall (accuracy)87.7
9thof 262GPT-5.4Overall (accuracy)87.5
10thof 262GPT-5.2Overall (accuracy)87.4
11thof 262Claude Sonnet 4.5Overall (accuracy) (thinking)87.4
12thof 262Claude Opus 4Overall (accuracy) (thinking)87.3
13thof 262Claude Opus 4.5Overall (accuracy) (thinking)87.3
14thof 262Claude Sonnet 4.6Overall (accuracy) (thinking)87.3
16thof 262GPT-5Overall (accuracy) (high)87.1
17thof 262Kimi K2.5Overall (accuracy)87.1
19thof 262Grok 4Overall (accuracy)87
20thof 262Seed-2.0-proOverall (accuracy)87
21stof 262Qwen3.5-122B-A10BOverall (accuracy)86.7
22ndof 262Seed 1.6Overall (accuracy) (thinking)86.6
25thof 262GPT-5.1Overall (accuracy)86.4

Read from the board on 2026-08-25