Mistral AI Launches OCR API, Beats Azure OCR, Google Gemini, and OpenAI GPT-4o

French AI company Mistral AI has unveiled Mistral OCR, a powerful new API for Optical Character Recognition that boosts document analysis. The tool processes images and PDFs, accurately pulling out structured text, media, tables, and equations.

“Approximately 90% of the world’s organisational data is stored as documents, and to harness this potential, we are introducing Mistral OCR,” said the Mistral AI. The API integrates with Retrieval-Augmented Generation (RAG) systems, making it suitable for processing multimodal documents such as slides and complex PDFs.

Mistral OCR is now the default model for document understanding on Le Chat and is available via the API ‘mistral-ocr-latest’ at 1000 pages per dollar, with batch inference doubling efficiency. 

The API is accessible on Mistral’s developer suite, La Plateforme, and will soon be available through cloud, inference partners, and on-premises deployment.

Mistral OCR supports multilingual and multimodal content, outperforming leading OCR models in benchmarks. It has been tested against Google Document AI, Azure OCR, Gemini models, and GPT-4o, scoring 94.89 overall, with high performance in mathematical expressions, scanned documents, and tables.

Mistral OCR  can handle a diverse range of scripts, fonts, and languages. “This versatility is crucial for both global organisations that handle documents from diverse linguistic backgrounds, as well as hyperlocal businesses serving niche markets,” the company said.

The API processes up to 2000 pages per minute on a single node. It also supports “doc-as-prompt” functionality, allowing structured output extraction in formats like JSON. This feature enables integration with downstream workflows.

Beta customers are using Mistral OCR for scientific research, historical preservation, customer service, and technical literature indexing. Research institutions have leveraged it to convert academic papers into AI-ready formats, while heritage organizations are digitizing historical records. Customer service teams are transforming manuals into searchable knowledge bases.

For enterprises handling sensitive data, Mistral AI offers a self-hosted deployment option. “Organisations with strict data privacy requirements can maintain full control over their infrastructure,” Mistral AI said.

Mistral AI plans to improve the model further and expand on-premises deployment in the coming weeks.