Mistral AI releases OCR 4 with bounding boxes, block classification, and confidence scores

Mistral AI released OCR 4, a document parsing system that returns bounding boxes, typed-block classification (titles, tables, equations, signatures, etc.), and inline confidence scores alongside extracted text. The model supports 170 languages across 10 language groups, runs in a single container for fully self-hosted deployments, and is available via API or Document AI in Mistral Studio. API pricing is $4 per 1,000 pages, with a 50% Batch-API discount reducing it to $2 per 1,000 pages. Document AI costs $5 per 1,000 pages. In human preference evaluations on over 600 documents across 12+ languages, annotators preferred OCR 4 over leading competitors with an average win rate of 72%. On the public OlmOCRBench it scored 85.20, the top overall score among tested models. On OmniDocBench it scored 93.07, though the company cautions that automated benchmarks have known limitations. OCR 4 is also integrated as an ingestion component of the Mistral Search Toolkit for RAG and enterprise search.