Blog Post

Azure AI Foundry Blog
4 MIN READ

Unlocking Document Intelligence: Mistral OCR Now Available in Azure AI Foundry

Naomi Moneypenny's avatar
Naomi Moneypenny
Icon for Microsoft rankMicrosoft
Apr 09, 2025

Reveal the hidden potential of your documents with Mistral OCR, now available in Azure AI Foundry. This state-of-the-art OCR model transforms unstructured content into actionable insights with unmatched speed, precision, and multilingual versatility.

Every organization has a treasure trove of information—buried not in databases, but in documents. From scanned contracts and handwritten forms to research papers and regulatory filings, this knowledg...
Updated Apr 09, 2025
Version 4.0
artificial intelligence
azure ai foundry
computer vision
generative ai
model catalog
dpo's avatar
dpo
Copper Contributor
Apr 14, 2025

Same issue. null returned with a basic test like below : 

curl "https://{endpoint url}/v1/ocr" \

  -H "Content-Type: application/json" \

  -H "Authorization: Bearer {API_KEY}” \

  -d '{

    "model": "mistral-ocr-2503",

    "document": {

      "type": "document_url",

      "document_url": "https://arxiv.org/pdf/2501.00663"

    }

  }'

  • dmusil's avatar
    dmusil
    Copper Contributor
    Apr 15, 2025

    dpo check this conversation on GitHub (nevermind the project this conversation is in). The comment left by user 'lfelder' has valuable info on how to at least getting OCR working on base64 encoded images. I can confirm this works.

    url = url_to_azure_endpoint
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer your_bearer_token"
}
base64_image = your_base_64_encoded_image
data = {
    "model": "mistral-ocr",
    "document": {
        "image_url": f"data:image/png;base64,{base64_image}"
    },
# Optional parameters:
    "include_image_base64": True,
    "image_limit": 5,
    "image_min_size": 100
}
response = requests.post(url, headers=headers, json=data)

    Microsoft still has to provide documentation on how to process documents via 'document_url' though as this seems to require some preprocessing we don't know about yet.