Bug: Deployed gpt-4o-transcribe models via Azure AI Foundry started returning error 500

[hantusk]

Bug: Deployed gpt-4o-transcribe models via Azure AI Foundry started returning error 500.

Tested across two different Azure accounts. Broke our production system all day today.

Fails in official @openai/openai npm package both in latest version 4.x.x and 5.x.x. The issue is with Azure AI Foundry's openai model deployment and not the javascript library.

Model endpoint returns error code 500 when including "model" as part of the request. Calling the API without this, makes it work. The three different model endpoints pointing to gpt-4o-transcribe all started failing today. First endpoint failed 2025-06-20 at around 9am CET.

--form-data-boundary-678cwr845j3mky88
Content-Disposition: form-data; name="model"

gpt-4o-transcribe
--form-data-boundary-678cwr845j3mky88
Content-Disposition: form-data; name="prompt"

REDACTED PROMPT
--form-data-boundary-678cwr845j3mky88
Content-Disposition: form-data; name="file"; filename="f62160b7fda86fa559fffda1766f5bc9.webm"
Content-Type: application/octet-stream

�Eߣ�B���B���B��B��B��webmB���B����S�g���������I�f�*ױ��B@M��ChromeWA�

Calling the endpoint manually without the model parameter returns a transcription response.

[leestott]

Hi did this get resolved there was a issue see https://learn.microsoft.com/en-ca/answers/questions/2285990/error-in-transcribing-audio-file-with-gpt-4o-trans

[hantusk]

Yes, thanks. I would love to get some reassurance that this would be visible through status.azure.com or another status page we could monitor if it should happen again