CtrlK
Mistral OCR 3 is a document-understanding and optical character recognition model from Mistral AI. Mistral's public governance pages list OCR 3 as an active specialized model family with a release date of December 15, 2025, while the official launch article presents it as a major upgrade over Mistral OCR 2.[1][2]
Mistral says the model is built to extract both text and embedded images from a wide range of documents and return markdown enriched with HTML table structure so downstream systems can preserve layout as well as text.[2]
| Area | Details |
|---|---|
| Release date | December 15, 2025 |
| Public model name | mistral-ocr-2512 |
| Main output style | Markdown with HTML table reconstruction |
| Main interfaces | API and Document AI Playground in Mistral AI Studio |
| Lifecycle status | Active |
Mistral's launch article says OCR 3 achieved a 74% overall win rate over OCR 2 on forms, scanned documents, complex tables, and handwriting.[2]
The company highlighted four upgrade areas:
| Area | What Mistral emphasized |
|---|---|
| Handwriting | Better reading of cursive and handwritten annotations |
| Forms | Better detection of boxes, labels, and dense layouts |
| Scanned documents | More robust to skew, noise, low DPI, and compression artifacts |
| Complex tables | Stronger reconstruction of merged cells and hierarchical headers |
Mistral priced OCR 3 at $2 per 1,000 pages, with a 50% Batch API discount that reduces the effective batch price to $1 per 1,000 pages.[2]
The company also said OCR 3 powers Document AI Playground in Mistral AI Studio, is fully backward compatible with OCR 2, and can be self-hosted for customers with strict data-privacy requirements.[2]
Mistral recommends OCR 3 for invoice and form extraction, enterprise document pipelines, archive digitization, and converting scientific or technical documents into text or structured JSON.[2]