Wikibench

MMLU-ProX

math

Extended version of MMLU-Pro providing additional challenging multiple-choice questions for evaluating language models across diverse academic and professional domains. Built on the foundation of the Massive Multitask Language Understanding benchmark framework.

Leaderboard

Showing 20 of 30 results

  1. Qwen3.7 Max
    87.0%
    i
  2. Qwen3.5-397B-A17B
    84.7%
    i
  3. Qwen3.6 Plus
    84.7%
    i
  4. Qwen3.5-122B-A10B
    82.2%
    i
  5. Qwen3.5-27B
    82.2%
    i
  6. Qwen3-235B-A22B-Thinking-2507
    81.0%
    i
  7. Qwen3.5-35B-A3B
    81.0%
    i
  8. Qwen3 VL 235B A22B Thinking
    80.6%
    i
  9. Qwen3-235B-A22B-Instruct-2507
    79.4%
    i
  10. Nemotron 3 Super (120B A12B)
    79.4%
    i
  11. Qwen3-Next-80B-A3B-Thinking
    78.7%
    i
  12. Qwen3 VL 235B A22B Instruct
    77.8%
    i
  13. Qwen3 VL 32B Thinking
    77.2%
    i
  14. Qwen3-Next-80B-A3B-Instruct
    76.7%
    i
  15. Qwen3.5-9B
    76.3%
    i
  16. Qwen3 VL 30B A3B Thinking
    76.1%
    i
  17. Qwen3 VL 32B Instruct
    73.4%
    i
  18. Qwen3.5-4B
    71.5%
    i
  19. Qwen3 VL 30B A3B Instruct
    70.9%
    i
  20. Qwen3 VL 8B Thinking
    70.7%
    i