Wikibench

MMStar

reasoning

MMStar is an elite vision-indispensable multimodal benchmark comprising 1,500 challenge samples meticulously selected by humans to evaluate 6 core capabilities and 18 detailed axes. The benchmark addresses issues of visual content unnecessity and unintentional data leakage in existing multimodal evaluations.

Leaderboard

Showing 20 of 22 results

  1. Qwen3.6 Plus
    83.3%
    i
  2. Qwen3.5-122B-A10B
    82.9%
    i
  3. Qwen3.5-35B-A3B
    81.9%
    i
  4. Qwen3.6-27B
    81.4%
    i
  5. Qwen3.5-27B
    81.0%
    i
  6. Qwen3 VL 32B Thinking
    79.4%
    i
  7. Qwen3 VL 235B A22B Thinking
    78.7%
    i
  8. Qwen3 VL 235B A22B Instruct
    78.4%
    i
  9. Qwen3 VL 32B Instruct
    77.7%
    i
  10. Qwen3 VL 30B A3B Thinking
    75.5%
    i
  11. Qwen3 VL 8B Thinking
    75.3%
    i
  12. Qwen3 VL 4B Thinking
    73.2%
    i
  13. Qwen3 VL 30B A3B Instruct
    72.1%
    i
  14. Qwen3 VL 8B Instruct
    70.9%
    i
  15. Qwen2.5 VL 72B Instruct
    70.8%
    i
  16. Qwen3 VL 4B Instruct
    69.8%
    i
  17. Qwen2.5 VL 32B Instruct
    69.5%
    i
  18. Qwen2.5-Omni-7B
    64.0%
    i
  19. Qwen2.5 VL 7B Instruct
    63.9%
    i
  20. DeepSeek VL2
    61.3%
    i