Wikibench

MMMU-Pro

reasoning

A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.

Leaderboard

Showing 20 of 56 results

  1. Gemini 3.5 Flash
    83.6%
    i
  2. GPT-5.5
    83.2%
    i
  3. Gemini 3 Flash
    81.2%
    i
  4. GPT-5.4
    81.2%
    i
  5. Gemini 3 Pro
    81.0%
    i
  6. Gemini 3.1 Pro
    80.5%
    i
  7. Muse Spark
    80.4%
    i
  8. Kimi K2.6
    80.1%
    i
  9. GPT-5.2
    79.5%
    i
  10. Qwen3.6 Plus
    78.8%
    i
  11. Kimi K2.5
    78.5%
    i
  12. GPT-5
    78.4%
    i
  13. MiniMax M3
    78.1%
    i
  14. MiMo-V2.5
    77.9%
    i
  15. Claude Opus 4.6
    77.3%
    i
  16. Gemma 4 31B
    76.9%
    i
  17. Qwen3.5-122B-A10B
    76.9%
    i
  18. Gemini 3.1 Flash-Lite
    76.8%
    i
  19. GPT-5.4 mini
    76.6%
    i
  20. o3
    76.4%
    i