CC-OCR.

Benchmark for OCR across multi-scene, multilingual, and document parsing tasks.

§ 01 · Multi-Scene F1

Multi-Scene F1.

F1 score on multi-scene text reading

Higher is better

#	Model	Score	Source
★	gemini-15-pro Non-API entry from src	83.25%	src
2	qwen2-vl-72b Non-API entry from src	77.95%	src
3	internvl2-76b Non-API entry from src	76.92%	src
4	gpt-4o Non-API entry from src	76.4%	src
5	claude-35-sonnet Non-API entry from src	72.87%	src

§ 02 · Multilingual F1

F1 score on multilingual text (10 languages)

Higher is better

#	Model	Score	Source
★	gemini-15-pro Non-API entry from src	78.97%	src
2	gpt-4o Non-API entry from src	73.44%	src

§ 03 · KIE F1

F1 score on key information extraction

Higher is better

#	Model	Score	Source
★	qwen2-vl-72b Non-API entry from src	71.76%	src
2	gemini-15-pro Non-API entry from src	67.28%	src
3	claude-35-sonnet Non-API entry from src	64.58%	src
4	gpt-4o Non-API entry from src	63.45%	src

§ 04 · Document Parsing

Average score on document parsing

Higher is better

#	Model	Score	Source
★	gemini-15-pro Non-API entry from src	62.37	src

§ Related · Explore