Mistral AI has launched OCR 4, a new tool designed to read text from documents such as PDFs, Word files, and PowerPoint presentations. Unlike previous versions, this model identifies the specific role of each element on a page, distinguishing titles, tables, equations, and signatures to break documents into meaningful sections. The system also provides confidence scores to indicate certainty about each word or page it processes. In a blind test involving over 600 documents, independent reviewers preferred the results 72 percent of the time compared to competing models. The model supports 170 languages and is accessible via the API, Mistral Studio, and Microsoft Foundry. Pricing stands at $4 per 1,000 pages, dropping to $2 when using batch mode.
This update matters because standard text extraction often fails to understand document structure, which complicates downstream tasks like searching or automated processing. By classifying blocks and offering confidence metrics, the tool reduces the need for manual cleaning before data enters other systems. The high preference rate in blind tests suggests the model handles diverse layouts and languages more reliably than current alternatives.
* Available through Mistral Studio and Microsoft Foundry APIs
* Costs $4 per 1,000 pages or $2 in batch mode
* Supports 170 languages including less common ones
