Product Launches

Mistral OCR 4.1 fixes bounding box drift on dense pages for precise document search

Share
Mistral OCR 4.1 fixes bounding box drift on dense pages for precise document search

Mistral has released OCR 4.1, an update to its OCR 4 model that fixes bounding box drift on dense, annotated pages. The update ensures precise alignment of bounding boxes to every element, eliminating nested images and improving document search reliability.

TL;DR

  • Mistral OCR 4.1 improves bounding box precision on dense, annotated pages, ensuring reliable document search and citation.
  • The update eliminates bounding box drift and nested images, maintaining the same pricing, benchmarks, and language support as OCR 4.
  • OCR 4.1 is automatically available to users of the mistral-ocr-latest alias, with no code changes required.

What happened

On August 13, 2026, Mistral released OCR 4.1, an update to its OCR 4 model. The update focuses on improving bounding box precision on dense, annotated pages, such as contracts, medical reports, and scientific papers.

OCR 4.1 eliminates bounding box drift and nested images, ensuring that each bounding box aligns precisely with its corresponding element. This improvement enhances the reliability of document search and citation.

The update maintains the same pricing, benchmarks, and language support as OCR 4. Users of the mistral-ocr-latest alias will automatically receive the updated model without any code changes.

Why it matters

Accurate bounding boxes are crucial for reliable document search and citation. OCR 4.1's improvements enable precise visual citations, making it an essential tool for developers and startups building document search and agentic pipelines.

The update's focus on real-world documents, such as contracts and medical reports, makes it particularly valuable for industries that rely on accurate document processing.

OCR 4.1's maintenance of the same pricing and language support as OCR 4 ensures that users can benefit from the improvements without incurring additional costs or compatibility issues.

Key facts

  • Mistral OCR 4.1 was released on August 13, 2026, as an update to the OCR 4 model.
  • The update improves bounding box precision on dense, annotated pages, eliminating drift and nested images.
  • OCR 4.1 maintains the same pricing as OCR 4: $4 per 1,000 pages for standard mode, $2 per 1,000 pages for batch API, and $5 per 1,000 pages for Document AI mode.
  • The model supports 170 languages across 10 language families.
  • OCR 4.1 is automatically available to users of the mistral-ocr-latest alias, with no code changes required.
  • The model is accessible through the Mistral API, Document AI inside Mistral Studio, and cloud platforms like Amazon SageMaker and Microsoft Foundry.
  • OCR 4.1 can be deployed in a single container for on-premise use, ensuring data confidentiality.
  • The model scored 85.20 on OlmOCRBench and 93.07 on OmniDocBench, with an average human preference rate of 72 percent over leading competitors.

Context

Mistral's OCR 4.1 update is part of a broader trend in the AI industry towards improving document processing capabilities. Accurate bounding boxes are essential for reliable document search and citation, making OCR 4.1 a valuable tool for developers and startups building document search and agentic pipelines.

The update's focus on real-world documents, such as contracts and medical reports, highlights the growing demand for accurate document processing in various industries. As businesses increasingly rely on digital documents, the need for precise and reliable OCR tools becomes more critical.

Mistral's commitment to maintaining the same pricing and language support as OCR 4 ensures that users can benefit from the improvements without incurring additional costs or compatibility issues. This approach underscores Mistral's focus on providing accessible and practical AI solutions for businesses of all sizes.

Topics

Join the discussion

Have a take on this story? Weigh in with our community on Facebook.

💬 Discuss on Facebook →