typhoon-ai/typhoon-ocr-7b
Typhoon-OCR-7B by typhoon-ai is a 7 billion parameter, bilingual document parsing model based on Qwen2.5-VL-Instruction, designed for real-world documents in Thai and English. It excels at extracting structured and layout-heavy information from various document types, including financial reports, academic papers, receipts, and infographics, outputting content in Markdown or HTML. The model features multi-layered interpretation for figures and outperforms GPT-4o and Gemini 2.5 Flash in Thai document understanding, particularly for complex layouts and mixed-language content.
Loading preview...
Typhoon-OCR-7B: Bilingual Document Parsing
Typhoon-OCR-7B is a 7 billion parameter vision-language model developed by typhoon-ai, specifically engineered for robust document parsing in both Thai and English. Inspired by models like olmOCR and built upon Qwen2.5-VL-Instruction, it specializes in extracting information from diverse real-world documents, including structured formats like financial reports and academic papers, as well as layout-heavy documents such as receipts and infographics.
Key Capabilities
- Bilingual Document Understanding: Optimized for complex layouts and mixed-language content in Thai and English.
- Structured Output: Generates Markdown for general text and HTML for tables, including complex layouts and merged cells.
- Multi-layered Figure Interpretation: Analyzes figures, charts, and diagrams through observation, context analysis, text recognition, and artistic/structural analysis, providing structured descriptions.
- Performance: Outperforms GPT-4o and Gemini 2.5 Flash in Thai document understanding benchmarks.
- Specific Prompting: Designed to work with predefined prompts for optimal performance, ensuring accurate extraction and formatting.
Good For
- Automated Document Processing: Ideal for businesses and researchers needing to extract data from a wide range of documents.
- Financial and Academic Analysis: Efficiently processes financial reports, academic papers, and government forms.
- Informal Document Handling: Capable of parsing receipts, menus, tickets, and infographics with layout-aware structures.
- Multilingual Data Extraction: Particularly strong for applications requiring high-accuracy OCR and information extraction from Thai and English documents.