An independent guide to PaddleOCR
paddleocr.dev is a community-written reference for developers who use PaddleOCR, the open-source OCR and document-parsing toolkit created by Baidu's PaddlePaddle team. It is not run by Baidu or the PaddleOCR maintainers.
What this site is for
The official documentation is thorough but spread over many pages and versions, and much of the material online still describes PaddleOCR 2.x, whose API no longer works. This site collects the parts people look up most: installing the right PaddlePaddle build, a working 3.x quick start, a 2.x to 3.x migration table, pipeline choice, the Python and CLI reference, serving and the MCP server.
What it is not
It is not the official PaddleOCR website, it does not host the software, and it does not offer support. For authoritative information use the GitHub repository and the official documentation. PaddleOCR is released under the Apache License 2.0.
How we keep the content accurate
Verified against primary sources
Install commands, parameters, CLI flags and version numbers are checked against the official documentation, the PyPI package metadata and the source repository. Code samples use the current 3.x API.
Figures are attributed
Accuracy and speed numbers come from Baidu's own publications and are labelled "reported by Baidu". We do not publish independent benchmark results we have not run, and we test your documents before you rely on any claim.
Dated and corrected
Pages show a last-reviewed date and are rechecked when a new PaddleOCR release ships. Spotted a mistake? Open an issue or discussion on the PaddleOCR community and mention paddleocr.dev, or use the contact route listed in the privacy policy.
Comparisons with other OCR tools aim to be fair: each tool has strengths, and the right choice depends on your documents, languages and hardware. The site has no paid placements, and links to other tools are not sponsored.
PaddleOCR in brief
Dates below come from the published papers and PyPI release history.
PP-OCR v1 to v3
Baidu published the practical ultra-lightweight PP-OCR system in 2020, followed by PP-OCRv2 (2021) and PP-OCRv3 (2022). PP-StructureV2 for layout and table analysis appeared in late 2022.
PaddleOCR 3.0
PaddleOCR 3.0 introduced PP-OCRv5, PP-StructureV3 and a new pipeline-based API built on PaddleX, and was not backward compatible with 2.x. PaddleOCR-VL, a vision-language document parser, followed in the autumn.
3.4 to 3.7
PaddleOCR 3.4.0 shipped on January 29, 2026, 3.6.0 (PaddleOCR-VL-1.6) on May 28 and 3.7.0 (PP-OCRv6 as the default model) on June 11. See the changelog.
Where the name comes from
PaddleOCR is built on PaddlePaddle ("PArallel Distributed Deep LEarning"), Baidu's open-source deep-learning framework.
Contribute to PaddleOCR
Bug reports, fixes and integrations go to the upstream project, which is maintained by Baidu engineers and open-source contributors.
PaddleOCR on GitHub →