Wikibench

ChartQA

reasoning

ChartQA is a large-scale benchmark comprising 9.6K human-written questions and 23.1K questions generated from human-written chart summaries, designed to evaluate models' abilities in visual and logical reasoning over charts.

Leaderboard

Showing 20 of 24 results

  1. Claude 3.5 Sonnet
    90.8%
    i
  2. Llama 4 Maverick
    90.0%
    i
  3. Qwen2.5 VL 72B Instruct
    89.5%
    i
  4. Nova Pro
    89.2%
    i
  5. Llama 4 Scout
    88.8%
    i
  6. Qwen2-VL-72B-Instruct
    88.3%
    i
  7. Pixtral Large
    88.1%
    i
  8. Mistral Small 3.2 24B Instruct
    87.4%
    i
  9. Qwen2.5 VL 7B Instruct
    87.3%
    i
  10. Nova Lite
    86.8%
    i
  11. DeepSeek VL2
    86.0%
    i
  12. GPT-4o
    85.7%
    i
  13. Llama 3.2 90B Instruct
    85.5%
    i
  14. Qwen2.5-Omni-7B
    85.3%
    i
  15. DeepSeek VL2 Small
    84.5%
    i
  16. Llama 3.2 11B Instruct
    83.4%
    i
  17. Phi-3.5-vision-instruct
    81.8%
    i
  18. Pixtral-12B
    81.8%
    i
  19. Phi-4-multimodal-instruct
    81.4%
    i
  20. DeepSeek VL2 Tiny
    81.0%
    i