AI BriefingMistral OCR 3 Released
Key point
Mistral has released Mistral OCR 3, which significantly enhances handwriting and complex table recognition performance.
Details
Mistral OCR 3 is a new OCR model that maximizes text and image extraction performance. Compared to the existing Mistral OCR 2, it achieves a 74% higher win rate in forms, scanned documents, complex tables, and handwriting recognition, demonstrating performance that surpasses existing enterprise and AI-native OCR solutions.
Key upgrades include the following:
- Handwriting: Enhanced recognition of cursive writing, mixed annotations, and handwriting on printed forms
- Forms: Improved detection of complex layouts and boxes/labels in invoices, receipts, government documents, etc.
- Scanned and Complex Documents: Achieved robustness in environments with compression noise, distortion, and low resolution (low DPI)
- Complex Tables: Perfectly reconstructs table structures including merged cells and hierarchies using HTML tags (
colspan,rowspan)
Developers can integrate the model via the API (mistral-ocr-2512), and can instantly convert PDFs and images into text or structured JSON through the Document AI Playground in Mistral AI Studio. The cost is very economical at $2 per 1,000 pages, and available at a 50% discount of $1 when using the Batch-API.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.