mistral-ocr-4-0 API limits

Alexander Novikov 40 Reputation points
2026-10-06T01:41:57.2333333+00:00

Good day, experts.

Mistral says that the limits for the new "mistral-ocr-4-0" model are 1,000 pages and 50MB.

However, when I deploy this model in Azure AI Foundry and use it via API, I appear to hit a limit of 30 pages.

Could you please confirm what are the actual limits for this model in Azure AI Foundry?

Kind regards,

Alex

Microsoft Foundry
Microsoft Foundry

A unified Azure platform for creating and managing AI models, agents, and applications with built‑in enterprise security, monitoring, and governance

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Andriy Bilous 12,356 Reputation points MVP
    2026-10-06T04:27:41.62+00:00

    Hello Alexander Novikov

    The 30-page behavior is expected for the current Azure-hosted mistral-ocr-4-0 deployment.

    For mistral-ocr-4-0 sold directly by Azure in Microsoft Foundry, Microsoft currently documents:

    • Maximum PDF pages: 30
    • Maximum PDF size: 30 MB
    • Model status: Preview

    This is a per-request model/service limit, not your subscription quota or RPM/TPM quota, so requesting a quota increase would not normally raise it. Microsoft Learn

    The confusion is understandable because Azure previously documented Mistral-OCR-2503 on the older Hub-based deployment path with 1,000 pages / 50 MB. That is a different model/deployment path and those limits should not be applied to the newer Azure-direct mistral-ocr-4-0. Microsoft Learn

    There are also recent Q&A cases for mistral-document-ai-2512, which uses the same current Azure document-processing limits, where Microsoft explicitly confirms 30 pages / 30 MB. Microsoft Learn

    So for your deployment the practical limit is:

    mistral-ocr-4-0 in Microsoft Foundry: ≤30 pages AND ≤30 MB per PDF request.

    For larger documents, split the PDF into chunks of no more than 30 pages/30 MB, process them separately, and merge the results while preserving the original page numbers. Mistral itself publishes a document-chunking utility specifically designed to work with provider-specific limits, including Azure MaaS endpoints. Mistral AI

    Mistral's native service and Azure Foundry are separate serving platforms, so the limits advertised for Mistral's own API do not automatically apply to an Azure-hosted deployment. Mistral's current documentation also describes document limits as service/account-specific rather than a universal model property. Mistral AI

    Official Microsoft reference: Foundry Models sold by Azure — Mistral models

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.