Mistral OCR 4.1

387 points · 154 comments on HN · read original →

Points and comments are a snapshot, not live.

Mistral releases OCR 4.1 with paragraph-level bounding boxes and block confidence scores.

Mistral AI launched OCR 4.1 in public preview on July 16, 2026. The model supports paragraph-level bounding box extraction, structural block labels, and block-level confidence scores. Pricing is $4 per 1000 pages for basic OCR, $5 per 1000 annotated pages. It is part of Mistral's Document AI stack and is available via API endpoints for OCR, annotations, and batching.

What commenters are saying

Commenters debate whether Mistral's role in the AI race is due to European regulation or legitimate data sovereignty concerns. Some argue regulations give Mistral a protected market niche, citing its 'Shieldstral' moderation model. Others counter that Chinese open-weight models can be run locally, making data sovereignty moot. A few defend Mistral's practical contributions like OCR and document extraction, noting Gemini is stronger in that domain. Several commenters reject the 'race' framing, arguing commoditization and open-source models make national leadership less meaningful.