Mistral has launched Mistral OCR 4, a new document understanding model that introduces structured output features such as bounding boxes, block classification, and inline confidence scores across 170 languages. This release is significant as it demonstrates leading performance on public benchmarks, including a top score of 85.20 on OlmOCRBench and an average preference rate of 72% in head-to-head evaluations against its competitors, particularly excelling in rare and low-resource languages. Developed for high-volume processing and equipped to handle various document types, OCR 4 can be integrated via API and utilized through platforms such as Mistral Studio and Microsoft Foundry, while also offering a self-hosted option to address data residency requirements.

Mistral: Mistral develops and deploys advanced AI models focused on language understanding and document processing. The company is releasing Mistral OCR 4 as a compact, self-hostable model that structures extracted content for downstream enterprise applications.
Document AI: Document AI is Mistral’s no-code interface within Mistral Studio that layers schema-based structuring and annotation on top of the OCR 4 model. It enables users to generate custom JSON outputs or annotated results without writing additional parsing code.
Mistral OCR 4: Mistral OCR 4 is a specialized document-understanding model that extracts text while providing bounding boxes, block classifications, and per-region confidence scores. It is positioned in the news as the core engine for structured ingestion in RAG pipelines, agent workflows, and compliance processes across 170 languages.
Amazon SageMaker: Amazon SageMaker is AWS’s managed platform for building, training, and deploying machine learning models at scale. In this release, it integrates Mistral OCR 4 to let customers run the model in their existing SageMaker environments without moving data.
Microsoft Foundry: Microsoft Foundry is Microsoft’s enterprise platform for AI model deployment and workflow integration. The news highlights its partnership with Mistral to make OCR 4 and Document AI available directly within Foundry for business users.
Mistral Search Toolkit: Mistral Search Toolkit is an open-source, composable search framework announced at the AI Now Summit for building RAG and enterprise search systems. OCR 4 serves as its structured ingestion component, supplying citation-ready blocks to the toolkit’s retrieval and evaluation pipeline.
Snowflake Parse Document: Snowflake Parse Document is an upcoming feature within the Snowflake data platform for parsing and structuring documents. It is scheduled to incorporate Mistral OCR 4 to support native document ingestion for analytics and retrieval workloads.

`json
{
“Platform Availability”: “Mistral OCR 4 and Document AI are accessible today via Mistral’s API, Mistral Studio, Amazon SageMaker, and Microsoft Foundry, with self-hosted deployment also offered for data-residency needs.”,
“Multilingual Performance”: “OCR 4 shows its widest accuracy gains on rare and low-resource languages compared with competing systems, supporting ten language groups and improving coverage where many OCR tools degrade.”,
“Structured Output Benefits”: “The model’s bounding boxes, block typing, and confidence scores enable source-grounded citations, redactions, semantic chunking for RAG, and human-in-the-loop verification in production document pipelines.”
}
`