CC-OCR: leaderboard

7,058 annotated images in four OCR tracks (scene text, multilingual text in 10 languages, document parsing, information extraction) across 39 subsets from Alibaba (2024); edit distance, TEDS and F1.

Source: arxiv.org.

Interactive version: theaggregate.ai/benchmark?slug=cc-ocr · How It Works · Data refreshed daily, snapshot 2026-09-05.