Mistral OCR

2025-03-06 · Publisher : Mistral AI · Domain : Artificial Intelligence

Mistral AI launches Mistral OCR, a document recognition API that extracts text, tables, equations and images from PDFs and images, with an overall score of 94.89% on its internal benchmark, at 1 dollar per 1000 pages.

Mistral AI launches Mistral OCR, a document recognition API that extracts text, tables, equations and images from PDFs and images, with an overall score of 94.89% on its internal benchmark, at 1 dollar per 1000 pages. ### What changes - API dedicated to document understanding: text, embedded images, tables, equations and complex layouts (LaTeX). - Multilingual and multimodal: thousands of scripts and fonts. - “Doc-as-prompt” mode and structured output (JSON). - Up to 2000 pages per minute on a single node; becomes the default document understanding model in Le Chat. ### Specifications - Inputs: images and PDFs; output in Markdown interleaving text and images. - License: proprietary; self-hosting available selectively for sensitive organizations. - Availability: La Plateforme, Le Chat; cloud and inference partners announced. - API identifiers: `mistral-ocr-latest`, `mistral-ocr-2503` (deprecated on 12/02/2025 according to the documentation). ### Pricing | Item | Price | |---|---| | OCR | 1 USD per 1000 pages | | Batch | about 2 times more pages per dollar | | Annotations | 3 USD per 1000 pages (archived pricing page, June 2025) | ### Benchmarks Scores published by the vendor (internal “text only” benchmark, Mistral OCR 2503): | Criterion | Mistral OCR | |---|---| | Overall | 94.89 | | Mathematics | 94.29 | | Multilingual | 89.55 | | Scanned documents | 98.96 | | Tables | 96.12 | | Multilingual (fuzzy match, average) | 99.02 | ### Key takeaways An accurate and inexpensive document extraction service, useful upstream of a RAG system on complex PDFs.

Tags : LLM, Mistral, OCR, Documents, API