Wikibench

OCRBench

vision official site →

Multilingual: no

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

Methodology

Imported from llm-stats public benchmark metadata. Modality: multimodal. Max score: 1. Categories: image_to_text, vision. Language: en. Verified by llm-stats: no.

Leaderboard

Showing 20 of 22 results

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