Mistral's OCR 4 wins 72% of blind tests — is document processing finally solved?
mistral released ocr 4, a model that parses pdfs and docs into structured elements and wins 72% of blind tests against competitors. this sets a new standard for automated document ingestion.
mistral ai unveiled ocr 4, a model designed to convert pdfs and office documents into structured data by identifying text, layout elements, and functional roles like signatures or tables. mistral claims the model defeated competitors in 72% of over 600 blind test cases and supports 170 languages with high accuracy. available via mistral studio and microsoft foundry at $4 per 1,000 pages, it directly targets the enterprise document automation market. this launch signals a shift from basic text extraction to semantic document understanding, potentially unlocking more advanced agent workflows.