Mistral OCR Release Notes

Mistral AI has introduced Mistral OCR, an Optical Character Recognition (OCR) API designed to transform complex digitized documents into structured, AI-ready formats. By extracting interleaved text and images with high precision, Mistral OCR enables the integration of multimodal documents—such as scientific papers, slides, and complex PDFs—into Retrieval-Augmented Generation (RAG) systems.

High-Accuracy Document Understanding

Mistral OCR provides state-of-the-art comprehension of complex document elements, including mathematical expressions, tables, interleaved imagery, and LaTeX formatting. The model is specifically designed to handle rich documents like scientific papers that contain a mix of charts, graphs, and equations, outputting the content as ordered interleaved text and images (e.g., into Markdown files).

Performance Benchmarks

Mistral OCR 2503 outperforms several leading OCR models across multiple dimensions of document analysis. In internal "text-only" tests using publication papers and web PDFs, Mistral OCR achieved the highest scores across all measured categories:

Model Overall Math Multilingual Scanned Tables
Mistral OCR 2503 94.89 94.29 89.55 98.96 96.12
Gemini-1.5-Flash-002 90.23 89.11 86.76 94.87 90.48
GPT-4o-2024-11-20 89.77 87.55 86.00 94.58 91.70
Azure OCR 89.52 85.72 87.52 94.65 89.52
Google Document AI 83.42 80.29 86.42 92.77 78.16

Multilingual Capabilities

Mistral OCR is natively multilingual, capable of parsing and transcribing thousands of scripts, fonts, and languages. It demonstrates superior "Fuzzy Match in Generation" performance (99.02) compared to Azure OCR (97.31), Gemini-2.0-Flash-001 (96.53), and Google-Document-AI (95.88).

Specific language benchmarks for Mistral OCR 2503 include:

  • Spanish (es): 99.54
  • German (de): 99.51
  • Portuguese (pt): 99.42
  • Italian (it): 99.42
  • French (fr): 99.20
  • Ukrainian (uk): 99.29
  • Russian (ru): 99.09
  • Hindi (hi): 97.55
  • Chinese (zh): 97.11
  • Turkish (tr): 97.00
  • Romanian (ro): 98.79

Technical Features and Deployment

Speed and Throughput

Mistral OCR is designed to be lightweight, allowing it to process up to 2,000 pages per minute on a single node.

Doc-as-Prompt and Structured Output

The API supports "doc-as-prompt," allowing users to provide specific instructions to extract information from documents into structured formats like JSON. This enables the chaining of extracted data into downstream function calls and the creation of AI agents.

Availability and Pricing

  • API: The mistral-ocr-latest model is available via la Plateforme at a cost of 1,000 pages per $1, with approximately double the pages per dollar available via batch inference.
  • Le Chat: Mistral OCR is the default model for document understanding for users of Le Chat.
  • Self-Hosting: On-premises deployment is available on a selective basis for organizations with highly sensitive or classified information.

Primary Use Cases

  • Scientific Research: Converting journals and papers into AI-ready formats to accelerate collaboration and workflows.
  • Cultural Heritage: Digitizing historical documents and artifacts for preservation and public access.
  • Customer Service: Transforming manuals and documentation into indexed knowledge bases to reduce response times.
  • Professional Literature: Converting engineering drawings, regulatory filings, and lecture notes into answer-ready formats.

Sources

Related

  • Dispatch
    Mistral OCR 3 Release Notes

    Mistral AI has released Mistral OCR 3, a high-fidelity document parsing model that achieves a 74% win rate over its predecessor on complex documents and handwriting, priced at $2 per 1,000 pages.

  • Dispatch
    Mistral OCR 4 release notes / what's new

    Mistral AI has released Mistral OCR 4, a high-performance document parsing model supporting 170 languages and providing structured outputs including bounding boxes, block classification, and confidence scores.

  • Dispatch
    Mistral OCR 4 Release Notes: SOTA Document Intelligence

    Mistral OCR 4 is a high-performance document extraction model supporting 170 languages and structured output including bounding boxes and block classification, deployable via API or self-hosted containers.

  • Dispatch
    Mistral OCR 4.1 launch – pricing, features, and community reaction

    Mistral AI released OCR 4.1 on July 16 2026, offering paragraph‑level bounding boxes, structural labels, and a €3.5 / 1000‑page price, sparking mixed reactions about speed, cost, and accuracy on HN.

  • Dispatch
    Mistral Agentic Search Release

    Mistral AI has introduced Agentic Search, a multi-step retrieval loop that enables AI models to navigate complex, dense documents and improve accuracy on financial and government data benchmarks.