Wikibench

CharXiv-D

reasoning

CharXiv-D is the descriptive questions subset of the CharXiv benchmark, designed to assess multimodal large language models' ability to extract basic information from scientific charts. It contains descriptive questions covering information extraction, enumeration, pattern recognition, and counting across 2,323 diverse charts from arXiv papers, all curated and verified by human experts.

Leaderboard

Showing 13 of 13 results

  1. Qwen3 VL 32B Instruct
    90.5%
    i
  2. Qwen3 VL 32B Thinking
    90.2%
    i
  3. GPT-4.5
    90.0%
    i
  4. GPT-4.1 mini
    88.4%
    i
  5. GPT-4.1
    87.9%
    i
  6. Qwen3 VL 30B A3B Thinking
    86.9%
    i
  7. Qwen3 VL 8B Thinking
    85.9%
    i
  8. Qwen3 VL 30B A3B Instruct
    85.5%
    i
  9. GPT-4o
    85.3%
    i
  10. Qwen3 VL 4B Thinking
    83.9%
    i
  11. Qwen3 VL 8B Instruct
    83.0%
    i
  12. Qwen3 VL 4B Instruct
    76.2%
    i
  13. GPT-4.1 nano
    73.9%
    i