PaddleOCR
lastmod 2025-10-28created 2025-10-202 edits
Turn any PDF or image document into structured data for your AI. A powerful, lightweight OCR toolkit that bridges the gap between images/PDFs and LLMs. Supports 100+ languages.
PaddleOCR-VL
- PaddleOCR-VL: this is all you need for document parsing - YouTube
- PaddleOCR-VL - Baidu releases a 0.9B ultra-compact vision-language model for multilingual OC | GeekNews
An ultra-compact vision-language model (VLM) optimized for document parsing that supports 109 languages including Korean and accurately recognizes complex elements such as formulas, tables, charts, and handwriting
- The core model, PaddleOCR-VL-0.9B, combines a NaViT-based dynamic-resolution visual encoder with the ERNIE-4.5-0.3B language model to achieve both high-precision recognition and fast inference
- A small but powerful VLM architecture that reaches the recognition performance of existing large models while keeping compute efficiency
- Achieves SOTA (state-of-the-art performance) on OmniDocBench and others, surpassing existing pipeline-based models
- Supports a wide range of writing systems, from Korean, Chinese, English, and Japanese to Russian, Arabic, Hindi, Thai, and more, so it can be used to automate document processing worldwide
- Its lightweight structure uses few GPU resources, and it can be easily deployed and integrated via Docker, CLI, and the Python API
- Outperforms 72B-class multimodal models on some items, a multilingual document processing solution ready for immediate use in real-world environments