Presentiamo

Mistral OCR 4 introduces advanced document analysis by providing bounding boxes, block classification, and inline confidence scores with extracted text. The model supports 170 languages and can be deployed self-hosted in a single container, ensuring data sovereignty and enabling cost-efficient, high-throughput processing. It outperforms major OCR systems in human preference evaluations and public benchmarks, serving as a key ingestion component for enterprise search, RAG, and agentic workflows. Developers can integrate the model via an API, while a no-code application called Document AI is also available for the same engine.