Mistral AI Launches OCR 4: Elevating Document AI with Unprecedented Accuracy and Enterprise Deployment

Mistral AI has released Mistral OCR 4, its latest optical character recognition model, engineered to enhance document understanding for enterprise applications. This new offering aims to provide robust capabilities for data extraction and processing, directly targeting the needs of advanced AI engineers and developers.

Core Capabilities and Language Support

Mistral OCR 4 introduces several key features for precise document analysis. It provides detailed bounding boxes for extracted elements, coupled with typed-block classification that identifies titles, tables, equations, and signatures. Crucially, it delivers inline confidence scores alongside the extracted text, offering transparency into the model's certainty for each output. The model boasts extensive multilingual support, covering 170 languages across 10 distinct language groups.

Deployment Flexibility and Enterprise Integration

Designed for scalability and efficiency, OCR 4 supports deployment within a single container, facilitating fully self-hosted environments. This architecture is optimized for cost-efficient, high-throughput batch processing, making it suitable for large-scale data ingestion. Its primary role is to serve as a foundational ingestion component for critical enterprise systems, including enhanced enterprise search, Retrieval Augmented Generation (RAG) pipelines, and various domain-specific retrieval frameworks. The model is compatible with common enterprise document formats, such as PDF, DOC, PPT, and OpenDocument.

Performance Benchmarks

Mistral OCR 4 demonstrates significant performance gains across multiple benchmarks. In evaluations by independent annotators, OCR 4 was preferred over every leading OCR and document-AI system tested, achieving an average win rate of 72%. It secured the top overall score of 85.20 on the public OlmOCRBench and led Mistral's internal Crawl Multilingual evaluation with a score of 0.98. Furthermore, it recorded a score of 93.07 on OmniDocBench, underscoring its accuracy and reliability.

Pricing and Ecosystem Integration

For API usage, Mistral OCR 4 is priced at $4 per 1,000 pages. A 50% discount is available for batch API processing, reducing the cost to $2 per 1,000 pages. The same core engine powers Document AI within Mistral Studio, priced at $5 per 1,000 pages. This new OCR solution is also integrated as an ingestion component of the Mistral Search Toolkit, an open-source, composable search framework currently in public preview, further extending its utility within Mistral AI's growing ecosystem.