Mistral launches OCR 4 document model

Mistral OCR 4 represents a significant shift from raw text transcription to structured document layout parsing, making it a powerful foundation for enterprise RAG and agentic workflows.

• –The model achieves top-tier results on OlmOCRBench (85.20) and OmniDocBench (93.07), outperforming enterprise solutions like Google Document AI and Azure OCR.
• –Paragraph-level bounding box localization and block typing (titles, equations, signatures) directly address the lack of structural metadata in previous OCR engines.
• –Native support for 170 languages maintains high transcription accuracy on low-resource and specialized scripts where competitors degrade.
• –With a single-container deployment option, enterprises can self-host high-volume document ingestion pipelines to satisfy strict data sovereignty requirements.
• –Priced at $4 per 1,000 pages (and $2 with the Batch API), it offers a highly cost-efficient alternative to general-purpose multimodal LLM document parsing.