Wikibench

AI2D

reasoning

AI2D is a dataset of 4,903 illustrative diagrams from grade school natural sciences (such as food webs, human physiology, and life cycles) with over 15,000 multiple choice questions and answers. The benchmark evaluates diagram understanding and visual reasoning capabilities, requiring models to interpret diagrammatic elements, relationships, and structure to answer questions about scientific concepts represented in visual form.

Leaderboard

Showing 20 of 32 results

  1. Claude 3.5 Sonnet
    94.7%
    i
  2. Qwen3.6 Plus
    94.4%
    i
  3. GPT-4o
    94.2%
    i
  4. Pixtral Large
    93.8%
    i
  5. Qwen3.5-122B-A10B
    93.3%
    i
  6. Mistral Small 3.2 24B Instruct
    92.9%
    i
  7. Qwen3.5-27B
    92.9%
    i
  8. Qwen3.6-35B-A3B
    92.7%
    i
  9. Qwen3.5-35B-A3B
    92.6%
    i
  10. Llama 3.2 90B Instruct
    92.3%
    i
  11. Llama 3.2 11B Instruct
    91.1%
    i
  12. Qwen3 VL 235B A22B Instruct
    89.7%
    i
  13. Qwen3 VL 32B Instruct
    89.5%
    i
  14. Qwen3 VL 235B A22B Thinking
    89.2%
    i
  15. Qwen3 VL 32B Thinking
    88.9%
    i
  16. Qwen2.5 VL 72B Instruct
    88.4%
    i
  17. Grok-1.5V
    88.3%
    i
  18. Qwen3 VL 30B A3B Thinking
    86.9%
    i
  19. Qwen3 VL 8B Instruct
    85.7%
    i
  20. Qwen3 VL 30B A3B Instruct
    85.0%
    i