PaddleOCR
Open-source OCR toolkit and document AI engine that converts PDFs and images into structured Markdown or JSON for LLM and RAG workflows.
At a Glance
Self-hosted PaddleOCR toolkit released under the Apache 2.0 license; no hosted service pricing is published.
Engagement
Available On
Listed Nov 2026
About PaddleOCR
PaddleOCR is an open-source OCR and document parsing toolkit from the PaddlePaddle organization, released under the Apache 2.0 license. It converts PDFs and images into structured, LLM-ready Markdown or JSON and supports text recognition in 100+ languages. The latest GitHub release listed is v3.7.0, published 2026-06-11.
What It Is
PaddleOCR combines two main capabilities: scene text recognition (the PP-OCR series) and document parsing (PaddleOCR-VL and PP-StructureV3). Users feed in scanned pages, photos, PDFs or office documents and receive structured output such as text, tables, formulas, charts and layout information. The project also offers an online Experience Center and APIs on its official website, plus local deployment through documented pipelines.
Models and Pipelines
- PaddleOCR-VL: a 0.9B vision-language model for document parsing. The README says PaddleOCR-VL-1.6 reaches 96.3% on OmniDocBench v1.6 and improves table, ancient document, rare character, seal and chart recognition.
- PP-StructureV3: converts complex PDFs and images to Markdown or JSON and provides finer-grained coordinates, including table cell and text coordinates.
- PP-OCRv6: a unified model covering 50 languages (Chinese, English, Japanese and 46 Latin-script languages) in tiny, small and medium tiers for edge, mobile and server use.
- HPD-Parsing: a high-throughput document parsing model with OpenAI-compatible serving or local inference through a customized vLLM runtime.
Deployment and Integration
Inference can run on Paddle static or dynamic graph, or Transformers backends, with ONNX conversion and acceleration through OpenVINO, ONNX Runtime or TensorRT. Hardware backends listed include NVIDIA GPU, Intel CPU, Kunlunxin XPU and other AI accelerators. Serving options let applications in C++, C#, Java and other languages call it, and PaddleOCR.js runs PP-OCRv5 in the browser. The README lists integrations with Dify, RAGFlow, Pathway and Cherry Studio.
Recent Updates
Version 3.5.0 added Word, Excel and PowerPoint to Markdown conversion and DOCX export of parsed results. Version 3.6.0 introduced PaddleOCR-VL-1.6, and 3.7.0 introduced PP-OCRv6, which the README says gives 5.2x CPU speedup with OpenVINO.
Community Discussions
Be the first to start a conversation about PaddleOCR
Share your experience with PaddleOCR, ask questions, or help others learn from your insights.
Pricing
Open Source (Apache 2.0)
Self-hosted PaddleOCR toolkit released under the Apache 2.0 license; no hosted service pricing is published.
- Apache License 2.0
- PaddleOCR-VL-1.6 document parsing model
- PP-StructureV3 structure-aware conversion to Markdown/JSON
- Supports 100+ languages
- Runs on CPU, GPU, XPU, NPU
Capabilities
Key Features
- PDF and image to Markdown/JSON conversion
- PaddleOCR-VL document parsing VLM
- PP-StructureV3 structure-aware parsing with coordinates
- PP-OCRv6 multilingual text recognition (50 languages in one model)
- 100+ language support
- Table, formula, chart and seal recognition
- Office document (Word, Excel, PowerPoint) to Markdown
- DOCX export of parsed results
- ONNX conversion and accelerated inference
- Multi-GPU and multi-process parallel inference
- Serving deployment for C++, C#, Java and other languages
- Browser inference via PaddleOCR.js
- Online Experience Center and APIs
