Azure OpenAI Responses API - File Upload capabilities

Question

When using the file upload capabilities for the responses API what file types are supported - is this only supported for PDF files?
When uploading these files - where are they stored and how can I access them? Can we get access to where they are being stored if they are being help in any Azure Blob storage or can I only retrieve these with the given file id?
Is the retention period for this storage the same as conversations and they are deleted after 30 days?

Answer 1

Hi Maxwell, Joanne,

The current behavior of file uploads with the Azure OpenAI Responses API can be a bit confusing, especially if you’re coming from Open Ai’s public platform.

Here’s how it works today, and what to check.

1. Supported file types For the Responses API in Azure OpenAI, the only documented supported file type for direct file input is PDF. Specifically, models with vision support can accept PDFs, where Azure OpenAI extracts text and page images internally. Other file types (like DOCX, PPTX, XLSX, CSV, etc.) are not documented as supported for Responses API file input at this time, even if they are supported on OpenAI’s public platform.

If you try to upload non‑PDF files, you may see inconsistent or unsupported behavior.

2. Where uploaded files are stored When you upload a file using the Files API for use with Responses (purpose set to assistants), Azure OpenAI returns a file_id.
The platform does not expose the underlying storage location (for example, you can’t browse or mount it as Azure Blob Storage). Access to the file is only via the API using that file_id. There’s no supported way to directly read or manage the backing storage yourself.

3. How to access the uploaded file You don’t “open” the file directly. Instead:

• Upload the PDF
• Capture the returned file_id
• Reference that file_id in your Responses API call

That’s the only supported access pattern today.

4. Retention and deletion The documentation does not currently define a fixed retention period for uploaded files in the Responses API. The file object includes an expires_at field, but there’s no published guarantee that it follows the same 30‑day lifecycle as conversations. In short, the exact retention window is not documented yet.

If lifecycle control is critical, it’s safest to:

• Treat uploaded files as temporary
• Re‑upload when needed
• Avoid assuming long‑term persistence

5. Troubleshooting tips If things aren’t working as expected:

• Confirm the file is a PDF
• Ensure the upload purpose is set to assistants
• Verify the model you’re using supports vision/PDF input
• Make sure you’re passing the returned file_id correctly in the Responses API request

Hope this helps clarify the current capabilities and limitations. Do let me know if you have any further queries.

Thankyou!

Answer 2

For the Azure OpenAI Responses API, the provided information only documents PDF support for file input.

1. Supported file types
   For the Responses API file input, the documented support is specifically for PDF files with models that support both text and image inputs. The guidance states that “models with vision capabilities support PDF input” and describes how PDFs are handled (text extraction plus page images). No additional file types are listed for Responses API file input in the provided material.
2. Where files are stored and how they are accessed
   When PDFs are uploaded via the file API for use with assistants/responses, the upload response includes a file object with fields such as:

      {
        "id": "assistant-KaVLJQTiWEvdz8yJQHHkqJ",
        "bytes": 4691115,
        "created_at": 1752174469,
        "filename": "nucleus_sampling.pdf",
        "object": "file",
        "purpose": "assistants",
        "status": "processed",
        "expires_at": null,
        "status_details": null
      }
   

   The documented usage pattern is:
   • Upload the PDF with purpose set to assistants (since user_data is not supported).
   • Use the returned id as a file_id when invoking the model.
   The context does not describe any direct access to the underlying storage location (such as Azure Blob Storage) for these uploaded files. Access is described only in terms of using the file_id with the API.
3. Retention period / deletion behavior
   The provided material does not state a retention period, nor does it confirm that uploaded files follow the same 30‑day deletion behavior as conversations. It only shows that the file object has an expires_at field but does not define how or when it is set or how long files are retained. No additional retention or lifecycle details for Responses API file uploads are present in the given context.

---

References:

• Use the Azure OpenAI Responses API
• Azure OpenAI Assistants Code Interpreter (Preview)
• Azure OpenAI On Your Data
• Azure OpenAI On Your Data (classic)
• Azure OpenAI frequently asked questions