tech
Mistral OCR 4: SOTA OCR for Document Intelligence
Today, we're releasing Mistral OCR 4, featuring bounding boxes, block classification, 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 serves as an ingestion component for enterprise search, RAG, and domain-specific retrieval pipelines. OCR 4 is a small, focused model, and this post covers what's new, how it performs on public and internal benchmarks, the known limitations of those benchmarks, and guidance on when to use the model API versus Document AI.
TL;DR
- Mistral OCR 4 introduces bounding boxes, block classification (titles, tables, equations, signatures), and inline confidence scores.
- The model supports 170 languages across 10 language groups, showing improved performance on specialized and low-resource languages.
- OCR 4 demonstrates breakthrough performance, preferred by independent annotators over leading systems and achieving top scores on public benchmarks like OlmOCRBench.
- It can be deployed in a single container for self-hosted solutions, ensuring data residency and compliance.
- OCR 4 serves as an ingestion component for enterprise search, RAG, and domain-specific retrieval pipelines, integrating with Mistral Search Toolkit.
- The API offers pure extraction mode or Document AI capabilities for structured JSON output, custom prompts, and image annotation.
- Pricing is $4 per 1,000 pages for the API, with a 50% discount for the Batch-API, and Document AI is $5 per 1,000 pages.