PaddleOCR is an OCR toolkit designed to convert PDF documents and images into structured data, supporting over 100 languages. It features advanced document parsing capabilities and integrates with various AI ecosystems.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
PaddleOCR is an OCR toolkit designed to convert PDF documents and images into structured data for AI applications. It supports over 100 languages and provides features for intelligent document parsing and universal text recognition.
How it works
PaddleOCR utilizes models like PaddleOCR-VL for document parsing, achieving high accuracy in recognizing text, formulas, and tables. It supports structured outputs in Markdown and JSON formats, enabling easy integration with AI systems. The toolkit is designed for efficient deployment across various hardware platforms, including CPUs and GPUs.
Getting started
To install PaddleOCR, use the following command:
pip install paddleocr
To perform OCR on an image, you can run:
paddleocr --image_dir ./doc/imgs/1.jpg
Recent releases
- v3.7.0 (2026-06-11): Introduced PP-OCRv6 with improved accuracy, achieving +4.6% detection and +5.1% recognition over PP-OCRv5_server.
- v3.6.0 (2026-05-28): Released PaddleOCR-VL-1.6, achieving over 96.3% accuracy on OmniDocBench v1.6.
- v3.5.0 (2026-04-21): Integrated with the Hugging Face ecosystem, supporting 20 major models.
- v3.4.1 (2026-04-14): Added support for AMD GPU and Intel Arc GPU hardware.
- v3.4.0 (2026-01-29): Released PaddleOCR-VL-1.5, achieving 94.5% accuracy on OmniDocBench.
Traction
PaddleOCR has 85258 stars and 11004 forks on GitHub, indicating a substantial level of engagement from the developer community.
Caveats
PaddleOCR is licensed under Apache-2.0. As of now, there are 226 open issues in the repository, and it was created on 2020-05-08.