Mistral AI’s documentation site has quietly posted a spec entry for OCR 4.1, model id mistral-ocr-4-1, listed as Public Preview under the Premier tier on a page dated July 16, 2026. The listing describes an update to Mistral’s Document AI stack. It extracts bounding boxes at the paragraph level, tags each detected block with a structural label, and attaches a confidence score to every block. That output lets a developer see exactly which extracted field maps to which part of a scanned document.
Pricing appears directly on the page: $4 per 1,000 pages for standard OCR runs, $5 per 1,000 pages when annotations are requested. OCR, bounding-box extraction, and structured annotations all share the /v1/ocr endpoint, while batching runs through a separate /v1/batch endpoint.
The page states a concrete per-page rate rather than making a pricing argument itself. That figure gives buyers a fixed number to weigh against whatever an incumbent cloud vendor charges for comparable document parsing, a comparison Mistral leaves for the reader to make.
Mistral AI’s documentation site published the OCR 4.1 spec on a page dated July 16, 2026.
