Model deployment returns OperationNotSupported after idle period

[Mordris]

I've encountered a recurring issue with a text-embedding-3-small deployment on Foundry. After a period of inactivity (a few hours), the model seems to enter a "deep sleep" state where it fails with a 400 BadRequestError, claiming the operation is not supported for that model—even though it is a dedicated embedding model.

The Error:

{
  "error": {
    "code": "OperationNotSupported",
    "message": "The embeddings operation does not work with the specified model, text-embedding-3-small. Please choose different model and try again."
  }
}

Observations:

• The deployment works perfectly upon initial setup.
• After sitting idle, it begins throwing the error above.
• The Workaround: If I "Edit" the deployment in the portal and click "Save" (without changing any parameters), the deployment immediately resumes working.

This suggests a potential issue with the deployment's lifecycle management or the routing metadata being lost during idle periods.

Logs for context:
Below is a trace showing the failure at 06:38 followed by a success at 07:05 immediately after performing a "ghost save" on the deployment settings.

# Failure at 06:38:42Z
BadRequestError: Error code: 400 - {'error': {'code': 'OperationNotSupported', 'message': 'The embeddings operation does not work with the specified model, text-embedding-3-small...'}}

# Success at 07:05:33Z (After manual re-save)
2026-04-13T07:05:34.335383Z [info] request [rag_service.app] duration_ms=676 method=POST path=/search/batch request_id=670eae02 status=200

Has anyone else experienced this behavior with the v3 embedding models on Foundry?

[0xsan-z]

Yes, we have also faced a similar issue with text-embedding-3-small . The error which we see is "ErrorCould not complete vectorization action. The vectorization endpoint returned status code '400' (BadRequest). Status: 502 (Bad Gateway)". Looks to be an issue at Microsoft End. Better raise a support ticket with them.