Introducing Mistral OCR 4

Mistral OCR 4 extracts and structures content from documents, providing bounding boxes, block classification, and inline confidence scores alongside the text. The model supports 170 languages and is compact enough for self-hosted deployment in a single container, addressing data residency and high-throughput batch processing needs. It is designed as an ingestion component for enterprise search, RAG, and domain-specific retrieval pipelines, integrating with the Mistral Search Toolkit. OCR 4 demonstrates strong performance on benchmarks, with its structured output enabling semantic chunking for RAG and providing primitives for AI agents.