Mistral OCR 4 on Prem: A structure-aware, containerized Document AI for regulated industries
techtimes.com

Mistral OCR 4 on Prem: A structure-aware, containerized Document AI for regulated industries

Tech News
1 min read

Published by AINave Editorial • Reviewed by Ramit

TL;DRMistral OCR 4 ships on June 23, 2026, offering on-prem structure-aware document understanding in a single Docker container with 170 language support, paragraph-level bounding boxes, and schema-driven JSON output. Pricing starts at $4 per 1,000 pages for the Standard API, $2 for Batch, and $5 for the Document AI tier.
Mistral AI released OCR 4 on June 23, 2026, introducing a structure-aware document understanding model that can run entirely within an organization’s own infrastructure. The model supports 170 languages and returns paragraph-level bounding boxes, typed-block labels, and per-word and per-page confidence scores. When a JSON schema is provided, OCR 4 output can be transformed into schema-conformant JSON via a lightweight Document AI step, all within the same API call. OCR 4 is designed to be deployed as a single Docker container, enabling data residency and sovereignty for regulated sectors such as healthcare, legal, financial services, and government. The release also ties OCR 4 into Mistral’s Search Toolkit, enabling seamless ingestion to retrieval-augmented generation pipelines without extra preprocessing layers. While benchmark figures are directionally useful, independent assessments emphasize testing OCR 4 on customers’ own documents due to artifacts in some evaluations. Pricing is $4 per 1,000 pages for the Standard API, $2 per 1,000 pages for Batch, and $5 per 1,000 pages for the Document AI tier that adds structured JSON output. A production webinar is scheduled for July 7, 2026, and Snowflake Parse Document integration is forthcoming.

Sources

Latest Tech News