LLM
22 June 2026
New Model PP-OCRv6: Universal OCR for 50 Languages
N
New-ZZZ desk
Hugging Face Blog · 1 month ago
PaddleOCR introduced PP-OCRv6—a new universal Optical Character Recognition (OCR) model that supports 50 languages and scales from 1.5 to 34.5 million parameters. The model is designed to solve real-world tasks, including documents, screenshots, and signs, and demonstrates a significant improvement in accuracy compared to previous versions. It offers a flexible architecture and integration with various backends, simplifying its use in different applications.
Why it matters
- —Improved Accuracy: The model shows a significant increase in text recognition and detection accuracy compared to previous iterations.
- —Multilingualism and Scalability: Support for 50 languages and three difficulty levels (tiny, small, medium) make it universal for global projects.
- —Practical Application: The model is optimized for real-world scenarios (screenshots, documents, signs) and has flexible integration with various systems (RAG, parsing).
Key facts
- PP-OCRv6 is the latest version of the universal PaddleOCR model for OCR.
- Supports 50 languages, including Chinese, Japanese, and 46 Latin languages.
- Showed an improvement in text detection accuracy of +4.6% and recognition of +5.1% compared to PP-OCRv5.
- The architecture includes RepLKFPN for detection and EncoderWithLightSVTR for recognition, enhancing efficiency across different text types.
The full text is in the original source. Here we provide a brief summary and key facts.
