Mistral launches OCR 4, turning document extraction into a full enterprise AI play

Insight summary

• •Mistral released OCR 4, a document AI model providing structured document representations with bounding boxes and block-type classification.
• •OCR 4 supports 170 languages, multiple document formats, and can run on enterprise infrastructure for sensitive data.
• •The model outputs localized text blocks with confidence scores for semantic mapping and traceability in compliance workflows.
• •Pricing starts at $4 per 1,000 pages with discounts available for batch processing.
• •Independent evaluations show a 72% human preference win rate and high benchmark scores, though Mistral advises cautious interpretation.
• •Early enterprise users report similar or better accuracy than competitors with significantly lower cost and faster processing.
• •The model is available via Mistral API, Mistral Studio, Amazon SageMaker, Microsoft Foundry, with Snowflake support soon.