Would you mind sharing how you got it working, i.e. which model and what your base url looks like? I’m trying for hours now to get it working and still no luck with latest 3.2.10 
All I did was go to cursor settings > Beta > Turn on Nightly Update Access. If it doesn’t work after that, try making a new chat and just sending “test”. Sometimes it works, sometimes it doesn’t though, not sure why. I still cannot upload images to the chat, it then says “Unauthorized User Openai API key” when its Azure and I don’t have an OpenAI key. Hope this helps!
Also try adding a custom model with the same name as your deployment (likely the same name as the model)
Hey, confirming the instructions from the post above. Quick version:
- Cursor Settings > Beta > Update Access > switch to Nightly, restart Cursor.
- Cursor Settings > Models > Add Model > enter the model name exactly as the Azure deployment name.
- Start a new chat, pick this model in the dropdown, send a test message.
A couple important notes:
- Nightly installs updates automatically, without asking, when you close Cursor. It’s a pre-release branch, so builds can be unstable. If you use one instance for production, it’s better to keep a separate one.
- This workaround isn’t 100% reliable for everyone. See Jacob_M’s post. Sometimes the first try fails, and starting a new chat helps.
- Image uploads in chat with Azure BYOK still fail with an OpenAI key error. That’s a separate issue.
The main bug with model is required is on Cursor’s side and is being tracked. I can’t share an ETA for a Stable fix yet.
Guys, the old GitHub - gabrii/Cursor-Azure-GPT-5: A service that allows Cursor to use Azure GPT-5 deployments. · GitHub is back and mantained, it works perfectly on my experience, even better than before since the new model extraction from metadata gives cursor the possibility to use tailored prompt for each openai model, instead of general ones.
In my experience this is truly almost perfect, with the only known bug with the images upload, that unfortunately isnt on our side and we can not do anything abt it
btw @deanrie , where should i report the bug abt the images? if its fine here:
So,basically when you upload an image to chat, cursor internal routing overrides the custom base URL with the openai default one, and this makes so:
1:the custom URL isnt respected
2:cursor gets api error of wrong api key(unless you have a valid openai api key as your custom baseurl key)
Thanks @riccardo.bevoni for taking over maintaining the project, and the great work on revamping it! 
Hey @riccardo.bevoni, thanks for the detailed write-up and for picking up proxy support. It really helps users while we’re fixing things.
On the image bug, it looks like images work here. This is the same issue @Jacob_M described above, and we’re already tracking it on our side. The root cause is basically what you said: for requests with images we use a hardcoded verification endpoint at api.openai.com, so the custom base URL gets ignored and the Azure key doesn’t work. I can’t share an ETA for a fix yet, but I’ll post an update in the thread as soon as I have one.
If you find anything else about Azure BYOK behavior, especially with specific request IDs, please drop it here. It helps a lot with debugging.
i have got a new problem to report, would love it to be fixed ASAP since its breaking for me and other users.
in short: the cursor internal routing doesnt route gpt 5.5 requests to the custom url anymore(does work with 5.4 tho for example), and returns error:“User API Key Rate limit exceeded”.
sorry if i retag if i already did, im not sure i did so already
Hey @riccardo.bevoni, thanks for the report. This looks like a separate bug from what we discussed above. Based on the User API Key Rate limit exceeded error, the request with gpt-5.5 is getting routed to our first-party deployment instead of your custom URL, and the rate limit is coming from our side. The fact that gpt-5.4 works with the same setup supports that.
To confirm and report this further, we need a few details:
- The Request ID from one of the failed
gpt-5.5requests Chat > top-right menu > Copy Request ID. - Your Cursor version Help > About.
- How exactly
gpt-5.5is added in Cursor Settings > Models. The model name, and whether it matches the deployment name on your side. - Whether it only happens on
gpt-5.5, or also on other recently added models too, likegpt-5.5-codex.
So we don’t mix it with the main thread about model is required since different symptoms are getting combined there, it’s better to open a separate thread for gpt-5.5 routing and share the info there. If you open one, drop the link here too and I’ll follow up.
Hi all!
Thanks for the reports. This issue should be fixed now. Please give it another try.