Mistral AI announced the release of Mistral OCR 3, a new document processing model engineered to extract text and structure from complex files.
The company claims the updated system achieves a 74% higher performance rate compared to its predecessor, specifically targeting difficult formats such as handwritten notes, dense forms, and distorted scans.
The model is available immediately via API under the identifier mistral-ocr-2512 and through a new drag-and-drop interface within the Mistral AI Studio.
The new architecture focuses on preserving document layout alongside raw text extraction.
It outputs markdown enriched with HTML-based table reconstruction, allowing the model to maintain column hierarchies and merged cells for downstream applications.
Mistral positioned the tool as a cost-effective alternative to existing enterprise solutions, pricing the service at $2 per 1,000 pages, with a discounted rate of $1 per 1,000 pages for batched processing.
This release coincides with the launch of “Document AI,” a user interface that parses files into structured JSON or text without requiring code.
The system is designed to feed data into generative AI pipelines, such as Retrieval-Augmented Generation (RAG) systems, by digitizing archives and operational documents.
The model supports all languages previously handled by the platform and is fully backward compatible with Mistral OCR 2.
Key Takeaways:
- Mistral AI released OCR 3 today to improve the extraction of text from handwriting, forms, and low-quality scans.
- The model outputs structured markdown with HTML table support and is priced at $2 per 1,000 pages.
- Developers can access the tool via API or the new no-code Document AI Playground interface.
You may also want to check out some of our other recent updates.
Wanna know what’s trending online every day? Subscribe to Vavoza Insider to access the latest business and marketing insights, news, and trends daily with unmatched speed and conciseness! 🗞️
