HunyuanOCR - Tencent's open source expert model for optical character recognition

What is HunyuanOCR

HunyuanOCR is a high-performance optical character recognition model open-sourced by the Tencent hybrid team, with only 1 billion references. Developed based on the hybrid multimodal architecture, using end-to-end design, can efficiently handle text detection, recognition and document parsing tasks. The model scored 94.1 points in the complex document test, surpassing mainstream products such as Google Gemini3-Pro, and supports translation in 14 small languages. The lightweight features are suitable for ticket recognition, video subtitle extraction and other scenes, open source address for GitHub and Hugging Face platform.

Features of HunyuanOCR

• Efficient Lightweight Architecture: 1B parameter count only, based on hybrid native multimodal architecture, significantly reduces deployment costs and is suitable for a wide range of hardware environments.
• End-to-end processing capabilityThe whole process from image input to result output can be processed end-to-end, and the optimal result can be achieved by single instruction and single inference, which is more efficient and convenient than the traditional solutions.
• Multi-language supportSupport for more than 100 languages, covering both monolingual and multilingual hybrid documents, adapting to globalized application scenarios.
• Full OCR capability: Cover classic OCR tasks such as text detection and recognition, complex document parsing, open field information extraction, video subtitle extraction, etc. with comprehensive features.
• Excellent performance: Achieve SOTA level in a number of core capabilities, such as complex document parsing, multi-scene text detection and recognition, etc., with leading performance.
• easy-to-use: Provide a concise interface and rich sample code , support for a variety of frameworks (such as vLLM, Transformers) , easy to get started and integration .

Core Benefits of HunyuanOCR

• Lightweight and efficientThe 1B parameter count is based on a highly efficient architectural design that significantly reduces deployment costs while maintaining high performance.
• end-to-end design: End-to-end processing from input image to output result without complex cascading, improving efficiency and accuracy.
• Multi-language supportSupport for more than 100 languages, covering both monolingual and multilingual hybrid documents, adapting to globalized application scenarios.
• superior performance: It reaches the SOTA level on tasks such as complex document parsing, multi-scene text detection and recognition, and is significantly ahead of similar models.
• easy-to-use: Provide concise API and rich sample code , support a variety of mainstream frameworks , easy to integrate and deploy .
• Wide range of application scenariosIt is suitable for document processing, ticket field extraction, video subtitle extraction, photo translation and many other scenarios.

What is the official website of HunyuanOCR

• Project website:: https://hunyuan.tencent.com/vision/zh?tabIndex=0
• Github repository:: https://github.com/Tencent-Hunyuan/HunyuanOCR
• Huggingface Model Library:: https://huggingface.co/tencent/HunyuanOCR
• Technical Report:: https://github.com/Tencent-Hunyuan/HunyuanOCR/blob/main/HunyuanOCR_Technical_Report.pdf
• Online Experience:: https://huggingface.co/spaces/tencent/HunyuanOCR

Who HunyuanOCR is for

• developers: Efficient, lightweight OCR solutions are needed to develop software and applications for document processing, image recognition, multilingual translation, and other functions.
• business user: Automated text extraction and translation tools are needed in areas such as document management, ticket processing, and content creation to improve productivity and quality.
• research worker: Multimodal research in areas such as natural language processing and computer vision requires powerful OCR tools to process image and text data.
• educator: The need to quickly extract and translate text content from literature and textbooks for teaching and research, and to support multilingual learning and research.
• content creator: In video production and graphic creation, it is necessary to extract textual information from images or perform multi-language translation to enrich content creation.
• regular user: The need to quickly translate or extract textual information from images in travel, study, office and other scenarios to improve life and work efficiency.