Wikibench

CC-OCR

multimodal

A comprehensive OCR benchmark for evaluating Large Multimodal Models (LMMs) in literacy. Comprises four OCR-centric tracks: multi-scene text reading, multilingual text reading, document parsing, and key information extraction. Contains 39 subsets with 7,058 fully annotated images, 41% sourced from real applications. Tests capabilities including text grounding, multi-orientation text recognition, and detecting hallucination/repetition across diverse visual challenges.

Leaderboard

Showing 18 of 18 results

  1. Qwen3.6 Plus
    83.4%
    i
  2. Qwen3 VL 235B A22B Instruct
    82.2%
    i
  3. Qwen3.6-35B-A3B
    81.9%
    i
  4. Qwen3.5-122B-A10B
    81.8%
    i
  5. Qwen3 VL 235B A22B Thinking
    81.5%
    i
  6. Qwen3.6-27B
    81.2%
    i
  7. Qwen3.5-27B
    81.0%
    i
  8. Qwen3 VL 30B A3B Instruct
    80.7%
    i
  9. Qwen3.5-35B-A3B
    80.7%
    i
  10. Qwen3 VL 32B Instruct
    80.3%
    i
  11. Qwen3 VL 8B Instruct
    79.9%
    i
  12. Qwen2.5 VL 72B Instruct
    79.8%
    i
  13. Qwen2.5 VL 7B Instruct
    77.8%
    i
  14. Qwen3 VL 30B A3B Thinking
    77.8%
    i
  15. Qwen2.5 VL 32B Instruct
    77.1%
    i
  16. Qwen3 VL 8B Thinking
    76.3%
    i
  17. Qwen3 VL 4B Instruct
    76.2%
    i
  18. Qwen3 VL 4B Thinking
    73.8%
    i