Wikibench

MathVision

math

MATH-Vision is a dataset designed to measure multimodal mathematical reasoning capabilities. It focuses on evaluating how well models can solve mathematical problems that require both visual understanding and mathematical reasoning, bridging the gap between visual and mathematical domains.

Leaderboard

Showing 20 of 28 results

  1. Kimi K2.6
    93.2%
    i
  2. Qwen3.6 Plus
    88.0%
    i
  3. Qwen3.5-122B-A10B
    86.2%
    i
  4. Qwen3.5-27B
    86.0%
    i
  5. Gemma 4 31B
    85.6%
    i
  6. Kimi K2.5
    84.2%
    i
  7. Qwen3.5-35B-A3B
    83.9%
    i
  8. Gemma 4 26B-A4B
    82.4%
    i
  9. Gemma 4 12B
    79.7%
    i
  10. Qwen3 VL 235B A22B Thinking
    74.6%
    i
  11. Step3-VL-10B
    70.8%
    i
  12. DiffusionGemma 26B-A4B
    70.5%
    i
  13. Qwen3 VL 32B Thinking
    70.2%
    i
  14. Qwen3 VL 235B A22B Instruct
    66.5%
    i
  15. Qwen3 VL 30B A3B Thinking
    65.7%
    i
  16. Qwen3 VL 32B Instruct
    63.4%
    i
  17. Qwen3 VL 8B Thinking
    62.7%
    i
  18. Qwen3 VL 30B A3B Instruct
    60.2%
    i
  19. Qwen3 VL 4B Thinking
    60.0%
    i
  20. Gemma 4 E4B
    59.5%
    i