Mistral OCR 4 (mistral.ai)

Mistral has launched OCR 4, an advanced optical character recognition model that introduces notable features such as bounding boxes, block classification, and inline confidence scores for extracted text. Supporting 170 languages across ten groups, this compact model is designed for fully self-hosted deployments, making it suitable for enterprise environments where data sovereignty is crucial. OCR 4 has demonstrated impressive performance, with independent assessments revealing a preference over leading OCR systems 72% of the time and achieving a top score of 85.20 on the OlmOCRBench, cementing its status as a capable tool for document parsing and retrieval. Significantly, OCR 4 enhances data extraction by not only localizing text but also categorizing content types, which facilitates better integration into retrieval-augmented generation (RAG) workflows and enterprise searches. The model’s structured output allows downstream systems to utilize precise location data and confidence scores, crucial for applications like compliance checks and invoice processing. Available via a straightforward API, OCR 4 is optimized for high-volume document processing while being cost-effective—priced at $4 per 1,000 pages—and is now integrated with Mistral's Search Toolkit, providing seamless ingestion into broader AI-driven workflows. Its ability to process rare and low-resource languages effectively further positions it as a versatile solution for diverse documentation needs.