Mistral OCR 4 : SOTA OCR for Document Intelligence

Mistral OCR 4 introduces bounding boxes, typed-block classification (titles, tables, equations, signatures), and inline confidence scores alongside extracted text. It supports 170 languages and can be deployed in a single container for self-hosted deployments. The model integrates with Mistral Search Toolkit for RAG and enterprise search pipelines. Benchmarks show it outperforms leading OCR and document-AI systems in human evaluations and on public benchmarks like OlmOCRBench and OmniDocBench, with specific improvements noted for specialized and low-resource languages. The post also details known limitations of automated benchmarks and recommends evaluation on user-specific documents.