We are seeing a consistent issue with the Responses API when passing a processed OpenAI file of type .docx (or pptx, xlsx …) as an input_file.
What works:
- Uploading the .docx to /v1/files succeeds
- OpenAI returns a valid file object
- The file status becomes processed
- The returned filename remains .docx
- We then pass that file_id into /v1/responses as an input_file
What fails:
- /v1/responses rejects the request with:
“The file type you uploaded is not supported. Supported extensions: … docx … xlsx … pptx …”
This appears inconsistent because:
- the upload step accepts the .docx
- the file is processed successfully
- the error message itself lists .docx as supported
- legacy .doc works in the same flow, but .docx fails
Environment:
- Model: gpt-4.1-mini
- We also reproduced similar behavior with gpt-5.4-mini
- File purpose: user_data
- File status before inference: processed
Reproduction steps:
- Upload a .docx file to /v1/files
- Receive file object successfully
- Wait until file status is processed
- Call /v1/responses with input containing:
- input_text
- input_file using that file_id
- Observe rejection from Responses API
Example observed metadata:
- original filename: test_file.docx
- mime type: application/vnd.openxmlformats-officedocument.wordprocessingml.document
- bytes: 13142
- returned file purpose: user_data
- returned status: processed
Example error:
“The file type you uploaded is not supported. Supported extensions: art, bat, … doc, docx, … pptx, … xlsx …”
Please confirm:
- Whether .docx is currently supported in the Responses API input_file path
- Whether support is model-dependent
- Whether there is a known issue with processed .docx files in /v1/responses
- Whether a different upload purpose or payload shape is required for .docx