Mistral OCR
lastmod 2026-08-13created 2025-03-073 edits
Mistral AI - Mistral OCR
About
- Mistral OCR is the world's best document understanding API, offering the ability to understand and analyze documents more accurately than existing models
- Extracts text, media, equations, and tables from PDFs and images and converts them into structured output
- The API is currently offered at 1000 pages/$1 (half the cost per page with batch processing)
Features
- Understanding of complex documents: accurately interprets tables, images, equations, and even LaTeX formatting
- Multilingual and multimodal support: supports many languages, fonts, and scripts
- Industry-leading performance: higher accuracy than other OCR models
- Top speed: can process 2000 pages per minute on a single node
- Documents can be used as prompts: supports structured output such as JSON
- On-premises (self-host) option: suitable for companies that want to process confidential documents