Mistral AI has officially released the new generation OCR technology, Mistral OCR 3, which achieves a major breakthrough in document processing accuracy and efficiency. Compared to the previous generation, Mistral OCR 3 has achieved a 74% overall win rate improvement in forms, scanned documents, complex tables, and handwriting recognition, surpassing enterprise-grade document processing solutions and AI-native OCR solutions in accuracy. As a more compact model, Mistral OCR 3 offers industry-leading cost-effectiveness, costing only $2 per 1000 pages processed, with batch API usage enjoying a 50% discount to $1. The model supports multiple document type outputs, including Markdown format with HTML table reconstruction, and can be used through API or the newly launched Document AI Playground interface. Mistral OCR 3 is particularly suitable for enterprise document processing pipelines, form and invoice parsing, historical document digitization, and other scenarios, providing high-quality document understanding capabilities for AI agents and knowledge systems.
