Mistral OCR 2

2025-05-22 · Publisher : Mistral AI · Domain : Artificial Intelligence

Mistral AI releases Mistral OCR 2 (mistral-ocr-2505) with document annotations: a new version of its OCR service, more reliable for extracting text and regions (bounding boxes) on varied documents.

What changes

  • New version of the OCR model (25.05) that succeeds mistral-ocr-2503.
  • Extraction of text and embedded images, with region extraction (bounding boxes).
  • Joint launch of structured annotations, which return fields extracted according to a schema.
  • No dedicated blog post: the announcement appears in the documentation changelog.

### Specifications - Inputs: images and PDFs; features: OCR, bounding box extraction, structured annotations, batch. - License: proprietary (Premier model). - Availability: La Plateforme API; version relayed on Google Cloud (Vertex AI) according to a Google Cloud post. - API identifier: `mistral-ocr-2505` (deprecated on 02/27/2026 according to the documentation). ### Pricing | Item | Price | |---|---| | OCR | 1 USD per 1000 pages | | Annotations | 3 USD per 1000 pages | Prices recorded from the pricing page archived on June 20, 2025 (a period when OCR 2 was the current model); exact pricing on May 22 not verified. ### Benchmarks No score published by the vendor for this version in the sources consulted. ### Key takeaways An incremental update to the OCR service, useful mainly for structured extraction through annotations in document pipelines.

Links

Tags : LLM, Mistral, OCR, Documents, API