I have used BYOK with Azure OpenAI for the last couple of days. It worked flawlessly. I used the latest Cursor Version available (3.5.33) and the model gpt-5.5 in Agent mode.
The issue is now, that since about 12 hours ago (May 26, 2026, 11:37 PM UTC), Cursor did not work anymore with this BYOK approach. Nothing has changed, the credentials are still correct and I can connect to Azure OpenAI via a normal program easily. The Cursor version also did not change.
However, I am now getting this error message, when I try to send a prompt using an Azure OpenAI key.
Network Error
We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.
I assume that something must have changed in the backend of Cursor that now makes these Azure OpenAI requests impossible.
I also tried out different models like gpt-5.4, but this did not work either.
Changing to HTTP/2 or to HTTP/1.1 also did not help.
Could you please look into this?
Steps to Reproduce
Select as model: gpt-5.5
Enter Azure OpenAI credentials.
Send a prompt in Agent mode.
Hey, this is a known BYOK bug for GPT-5.5, including the Azure version. I’ve reported it internally and it’s being tracked. The Network Error you’re seeing is a 502 on our side when we try to proxy the request, it’s not an issue with your Azure creds.
I can’t share an ETA for a fix yet. Once I have an update, I’ll reply in the thread.
As a temporary workaround, you can try a different model (Claude or Gemini via BYOK, or via Cursor inference) if that works for your task.
Thanks for the details, your point #4 is especially helpful.
This lines up with what we’re seeing. The requests reach our backend and fail there before they ever get sent to Azure. That’s why you get the same error with any credentials, and why changing the deployment name or version, including Nightly 3.6, doesn’t change anything. Your Azure creds and setup look fine, the issue is on our side.
Windows Update is most likely just a timing coincidence. The logs show this isn’t a client-side issue.
This is the known BYOK issue for GPT-5.5, including the Azure version, which I’ve already reported. I can’t share an ETA for a fix yet, but I’ll post an update here as soon as I have one.
For now, as a temporary workaround, try a different model like Claude or Gemini, either via BYOK or via Cursor inference, if that works for your use case.
I found the solution now. It was embarrassing enough for me that I felt the urge to delete my previous posts.
I misspelled the “Base URL” of the Azure OpenAI Endpoint.
That was all. I double checked everything, restarted, reinstalled but always overlooked one tiny character.
The only feedback I have here and something that could help prevent these kind of errors in the future for other users: Please adjust your error message. The error message, which I mention in my very first post was not very specific and even misleading.
If the combination of API Key and Base URL are not working, then it should be clearly communicated as such.
However, the fault was completely on my side and I apologize that you spent time looking into that.