Getting a 404 error when calling OpenAI gpt-realtime resource

Saul Santos 5

I deployed a new gpt-realtime model for realtime speech, and copy the provided "Endpoint" from the deployment info window, but everytime I try to connect I get the 404 error.

The provided "Endpoint" is of the form: https://MYRESOURCE.openai.azure.com/openai/realtime?api-version=2024-10-01-preview&deployment=gpt-realtime

I have other models deployed under the same resource and I can connect without issues to all of them by using a provided endpoint and key for each of those models.

I tried to use different "api-version" values as recommended somewhere else in "learn.microsoft.com", i.e., "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/realtime-audio?source=recommendations" but none of the different ones I saw works.

Any idea of what it could be happening? I would appreciate any help.

Anonymous

2025-09-22T18:40:24.9533333+00:00
Hello Saul Santos

A 404 error when connecting to a gpt-realtime deployment usually means the endpoint or connection method is incorrect. Realtime models require WebRTC or WebSocket, not standard REST calls. The correct endpoint format is:

wss://<your-resource>.openai.azure.com/openai/realtime?deployment=<deployment-name>&api-version=2024-10-01-preview

or for WebRTC:

https://<your-resource>.openai.azure.com/openai/realtime?deployment=<deployment-name>&api-version=2024-10-01-preview

Ensure the deployment parameter exactly matches your Azure deployment name, and include the header Authorization: Bearer <your-api-key>. The API version must be 2024-10-01-preview; older versions will fail. Also confirm your resource is in a Global Standard (PayGo) region, as gpt-realtime is not supported in all regions. Finally, do not use /chat/completions or /responses paths—those are for non-realtime models. If you follow these steps and use the correct protocol, the 404 error should be resolved.

Hope it helps!

Thank you
Saul Santos 5 Reputation points

2025-09-22T19:09:11.7533333+00:00

Thanks for the comment. I am using WebSockets, using and following the "Quickstart" example (https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-quickstart?tabs=windows%2Ckeyless&pivots=programming-language-python), even the code transforms any provided endpoint url to the correct wss protocol and structure, so it's not trying to use standard REST call. I even modified the internal framework's hard-coded "api_version" variable to test all other possible ones, but none of them worked.

Even Postman cannot connect using wss.
Saul Santos 5 Reputation points

2025-09-23T18:37:02.01+00:00

UPDATE: Suddenly connecting to the model started happening this morning. I did not change anything on the code and or configuration on the deployment. It just happened. I tested with different api versions as before and it works only with the "2025-04-01-preview".
Saul Santos 5 Reputation points

2025-09-24T16:15:29.3366667+00:00
Correction: it seems I forgot that I had to modify the "Azure_Voice" library under my current setup in order to make it work.

The change I did was in the file located at:

"\env\Lib\site-packages\azure\ai\voicelive\aio\_patch.py"

At line 691

From: path`` = parsed.path.rstrip("/") + "/voice-agent/realtime"

To: path`` = parsed.path.rstrip("/")

This is an issues on my setup that it may not be present in others, so I have included here in case anyone has the same issue. I tried different endpoint formats (including the suggested above) and different "api_version" parameters but nothing seemed to work until I made the change on the "_patch.py" file. As stated before, the API version that worked + this file change is "2025-04-01-preview".

I installed all libraries following this Quickstart. The following is a list of my virtual environment libraries, for your reference:

aiohappyeyeballs==2.6.1

aiohttp==3.11.18

aiosignal==1.4.0

annotated-types==0.7.0

anyio==4.10.0

attrs==25.3.0

azure-ai-voicelive==1.0.0b1

azure-core==1.35.0

azure-identity==1.22.0

certifi==2025.4.26

cffi==1.17.1

charset-normalizer==3.4.3

colorama==0.4.6

cryptography==44.0.3

distro==1.9.0

frozenlist==1.7.0

h11==0.16.0

httpcore==1.0.9

httpx==0.28.1

idna==3.10

isodate==0.7.2

jiter==0.11.0

msal==1.33.0

msal-extensions==1.3.1

multidict==6.6.4

numpy==2.2.5

openai==1.108.0

propcache==0.3.2

PyAudio==0.2.14

pycparser==2.22

pydantic==2.11.9

pydantic_core==2.33.2

PyJWT==2.10.1

python-dotenv==1.1.0

requests==2.32.3

six==1.17.0

sniffio==1.3.1

sounddevice==0.5.1

tqdm==4.67.1

typing-inspection==0.4.1

typing_extensions==4.13.2

urllib3==2.4.0

websocket-client==1.8.0

websockets==15.0.1

yarl==1.20.1

Hope this helps.

Your answer