Wikibench

OCRBench

vision

OCRBench: Comprehensive evaluation benchmark for assessing Optical Character Recognition (OCR) capabilities in Large Multimodal Models across text recognition, scene text VQA, and document understanding tasks

Leaderboard

Showing 20 of 22 results

  1. Kimi K2.5
    92.3%
    i
  2. Qwen3.5-122B-A10B
    92.1%
    i
  3. Qwen3 VL 235B A22B Instruct
    92.0%
    i
  4. Qwen3.5-35B-A3B
    91.0%
    i
  5. Qwen3 VL 30B A3B Instruct
    90.3%
    i
  6. Qwen3 VL 8B Instruct
    89.6%
    i
  7. Qwen3 VL 32B Instruct
    89.5%
    i
  8. Qwen3.5-27B
    89.4%
    i
  9. Qwen3.6-27B
    89.4%
    i
  10. Qwen2.5 VL 72B Instruct
    88.5%
    i
  11. Qwen3 VL 4B Instruct
    88.1%
    i
  12. Qwen2-VL-72B-Instruct
    87.7%
    i
  13. Qwen3 VL 235B A22B Thinking
    87.5%
    i
  14. Qwen2.5 VL 7B Instruct
    86.4%
    i
  15. Qwen3 VL 32B Thinking
    85.5%
    i
  16. Phi-4-multimodal-instruct
    84.4%
    i
  17. Qwen3 VL 30B A3B Thinking
    83.9%
    i
  18. DeepSeek VL2 Small
    83.4%
    i
  19. Qwen3 VL 8B Thinking
    81.9%
    i
  20. DeepSeek VL2
    81.1%
    i