Mistral OCR 3 (documentation: https://mistral.ai/fr/news/mistral-ocr-3 and demonstration: https://console.mistral.ai/build/document-ai/ocr-playground) is the latest character recognition model in Mistral.

Mistral OCR 3 accounts for the structure of the loaded text: images are identified, graphics are transcribed as such and columns are processed in reading order.
Text transcription or Markdown.

Accepts up to 5 documents and images (PDF, PNG, JPG, JPEG) of up to 8 pages and 10 MB each.

Accessible for free from “Documentary AI” in the Mistral Studio with an “Experiment” plan. No limitations observed during tests with short documents.

Test from Baudeaire’s manuscript “The Port” https://commons.wikimedia.org/wiki/File:Le_port_manuscrit_de_Baudelaire.jpg

Test from excerpts “The State of School 2025”, DEPP, https://www.education.gouv.fr/l-etat-de-l-ecole-2025-451742

Testing since https://www.education.gouv.fr/insertion-professionnelle-des-apprentis-de-niveau-cap-bts-six-mois-apres-leur-sortie-d-etudes-en-451984

Testing from https://commons.wikimedia.org/wiki/File:The_corn_law_question_shortly_investigated_(Hunter,_table,_p.22).jpg?uselang=en