Mistral AI has launched Mistral OCR 3, a new optical character recognition model that delivers a 74% overall win rate over its predecessor Mistral OCR 2 in extracting information from forms, scanned documents, complex tables, and handwritten content. This model is now powering the Document AI Playground within Mistral AI Studio, enabling users to drag and drop PDFs and images for conversion into clean text or structured JSON. The system is available to developers through an API, supporting markdown output with HTML-based table reconstruction, and priced at $2 per 1,000 pages with a 50% discount for batch processing.
We'd love for you to try the API or check out the Document AI playground! Both available on Mistral AI Studio. pic.twitter.com/20Pap6KblJ
— Mistral AI (@MistralAI) December 18, 2025
The release is immediately accessible to both developers and enterprise users globally through the Mistral AI Studio interface and API. Mistral OCR 3 targets organizations needing high-volume, accurate document parsing, such as those processing invoices, compliance forms, scientific reports, or digitizing handwritten archives. Its technical advances include robust handling of low-quality scans, dense layouts, and complex handwritten annotations. Compared to leading enterprise and AI-native OCR solutions, it offers higher accuracy and smaller model size, resulting in lower operational costs.
Early users are leveraging the tool for large-scale digitization and information extraction, with positive feedback from industry analysts who highlight the potential for increased data value. Mistral AI, known for its focus on advanced AI models for language and document understanding, continues to expand its product suite to address real-world business document challenges, aiming for seamless integration into existing enterprise pipelines and knowledge systems.
