Skip to content

MMLU-Redux

A harder refresh of MMLU intended to keep broad knowledge evaluation useful after the original benchmark became too easy for frontier models.

15 models

Higher is better.

  1. 1Claude Opus 4.5Anthropic96.6
  2. 2Qwen3.7 MaxAlibaba95.0
  3. 3Qwen3.5 397BAlibaba94.9
  4. 4Qwen3.6 PlusAlibaba94.5
  5. 5Qwen3.7 PlusAlibaba94.5
  6. 6Kimi K2.5Moonshot AI94.5
  7. 7Qwen3.6-27BAlibaba93.5
  8. 8QwQ-32BAlibaba90.0
  9. 9Qwen3 235B-A22BAlibaba89.2
  10. 10Ternary Bonsai 2 27BPrism ML89.1
  11. 11Mellum2.1-12B-A2.5B-ThinkingJetBrains87.8
  12. 12Qwen3 32BAlibaba85.7
  13. 13MiniCPM5-2BOpenBMB84.7
  14. 14Mellum2-12B-A2.5B-InstructJetBrains78.1
  15. 15MiniCPM5-1BOpenBMB70.1
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Multiple choice questions
Tasks
Broad academic QA
Authors
Qwen (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.