CC-OCR

Name: CC-OCR Benchmark Results
Creator: South China University of Technology
License: https://creativecommons.org/licenses/by/4.0/

South China University of Technology

Multi-scene text reading, key information extraction, multilingual text, and document parsing benchmark.

Models5

Papers12

Metrics4

SOTA History

Not enough data to show trend.

F1 score on multi-scene text reading

Higher is better

Rank	Model	Source	Score	Year	Paper
1	gemini-15-pro Multi-Scene Text Reading - Overall F1 score	Editorial	83.25	2025	Source
2	qwen2-vl-72b	Editorial	77.95	2025	Source
3	internvl2-76b	Editorial	76.92	2025	Source
4	gpt-4o	Editorial	76.4	2025	Source
5	claude-35-sonnet	Editorial	72.87	2025	Source

F1 score on multilingual text (10 languages)

Higher is better

Rank	Model	Source	Score	Year	Paper
1	gemini-15-pro Multilingual Text Reading - 10 languages	Editorial	78.97	2025	Source
2	gpt-4o	Editorial	73.44	2025	Source

F1 score on key information extraction

Higher is better

Rank	Model	Source	Score	Year	Paper
1	qwen2-vl-72b Key Information Extraction - Overall F1 score	Editorial	71.76	2025	Source
2	gemini-15-pro	Editorial	67.28	2025	Source
3	claude-35-sonnet	Editorial	64.58	2025	Source
4	gpt-4o	Editorial	63.45	2025	Source

Average score on document parsing

Higher is better

Rank	Model	Source	Score	Year	Paper
1	gemini-15-pro Document Parsing - Average Score	Editorial	62.37	2025	Source