Mistral has announced the launch of Mistral OCR 3, a new optical character recognition model that claims a 74% overall win rate over its predecessor on forms, scanned documents, complex tables, and handwriting. The model is described as state‑of‑the‑art, outperforming both enterprise document processing solutions and AI‑native OCR solutions. It can extract text and embedded images from a wide range of documents with exceptional fidelity, supporting markdown output enriched with HTML‑based table reconstruction so downstream systems can understand both content and structure. Mistral OCR 3 is available at an industry‑leading price of $2 per 1,000 pages, with a 50% batch‑API discount reducing the cost to $1 per 1,000 pages. Developers can integrate the model (mistral‑ocr‑2512) via API, and users can leverage a new Document AI Playground in Mistral AI Studio, a drag‑and‑drop interface that parses PDFs or images into clean text or structured JSON instantly. The release emphasizes improved performance on forms, handwritten content, low‑quality scans, and tables, and highlights robust handling of compression artifacts, skew, distortion, low DPI, and background noise. Mistral OCR 3 also reconstructs complex table structures with headers, merged cells, multi‑row blocks, and column hierarchies, outputting HTML table tags with colspan and rowspan to preserve layout. The model is backward compatible with OCR 2 and can be accessed via API or the new playground. For organizations with stringent data privacy requirements, Mistral offers a self‑hosting option to keep sensitive information within their own infrastructure. Early customers are using the model to process invoices into structured fields, digitize company archives, extract clean text from technical and scientific reports, and improve enterprise search. The release underscores that OCR remains foundational for enabling generative AI and agentic AI, with efficient, cost‑effective extraction of text and embedded images providing richer context and competitive advantage.
Introducing Mistral OCR 3
Mistral has released OCR 3, claiming a 74% overall win rate over its predecessor on forms, scanned documents, complex tables, and handwriting. The new model offers state‑of‑the‑art accuracy, supports markdown and HTML‑based table reconstruction, and is available at $2 per 1,000 pages, $1 with a batch discount. It can be accessed via API or a drag‑and‑drop Document AI Playground in Mistral AI Studio and is backward compatible with OCR 2. The release includes a self‑hosting option for data‑privacy‑constrained organizations.
Why it matters
High‑fidelity OCR is foundational for generative and agentic AI, enabling richer context and competitive advantage in data‑driven enterprises.
Topic dossiers Mistral AI