I get this We’re having trouble finding the resource you requested. If the problem persists, please contact support. Request ID: 29ba05e5-82f1-40ec-88e1-2c599b5e0f68, I can curl to the same end point without any issues. Suddely stop working yesterday.
Getting the same issue after updating last week
Previous versions didn’t have any problems with the same api and models
Azure OpenAi appears to be broken in the latest release?
This is a known issue on our side affecting Azure BYOK for the GPT‑5 family (GPT‑5.x) specifically, and it lines up with a change on our end around July 10‑11. Your curl works because the request Cursor builds for these models hits a different Azure path than a plain curl call. I’ve flagged both of your reports to the team.
A couple of ways to unblock yourself right now:
- Point your Azure key at a non‑GPT‑5 deployment (a GPT‑4.1 or GPT‑4o deployment) and pick it in the model picker under Cursor Settings > Models. Those use a different request path that isn’t affected, so they should work with your existing Azure setup.
- Or use a GPT‑5 model through Cursor’s built‑in access (without the custom Azure key). That routes fine - note it draws on your normal Cursor usage rather than your Azure account.
One thing that would really help us pin down your case: is your Azure OpenAI endpoint a direct *.openai.azure.com resource, or is it behind Azure API Management (APIM) or another gateway/proxy? And which exact model are you selecting when it fails?
Endpoint is a direct *.openai.azure.com Azure OpenAI resource — not behind APIM or any gateway/proxy.
Failures reproduce when selecting GPT-5.5 / GPT-5.6 Sol / Terra / Luna in Cursor with Azure BYOK enabled. The matching Azure deployments all return HTTP 200 via curl on both:
/openai/deployments/<deployment>/chat/completions/openai/v1/chat/completions?api-version=v1
Cursor fails at Object.classify with 400 invalid_input: model is required.
Cursor 3.11.19 Stable
This is a known issue on our side with Azure BYOK for the GPT‑5 family, and our team is actively working on a fix. Since your curl calls succeed, this is on us, not your Azure setup.
The model is required error you’re seeing looks a little different from the main failure we’ve been tracking, so I want to make sure your exact case is covered rather than assume it’s the same one. Could you grab the Request ID from one of the failing GPT‑5.5 / GPT‑5.6 attempts (three-dot menu on the message > Copy Request ID) and paste it here? Here’s how to find it. That lets us confirm exactly what we’re sending on these requests.
In the meantime, the quickest way to stay unblocked is to point your Azure key at a GPT‑4.1 or GPT‑4o deployment (those use a different request path that isn’t affected), or use a GPT‑5 model through Cursor’s built-in access without the custom key.
Request ID: 115549bb-9037-48f8-bf01-b62c6723d280
thank you about the help, about the unblocked part - gpt-4.1 is very bad and gpt-4o is something i don’t ever want to touch bro , but it’s ok we have other fallbacks 
Thanks for grabbing that, the request ID told us exactly what we needed. I traced it and confirmed this one’s on us: for GPT-5.5/5.6 on Azure BYOK we’re building the request without the model identifier, so it gets rejected on our end before it ever reaches your Azure endpoint. That’s why your curl calls (which carry the deployment in the path) go through fine, nothing wrong with your setup.
It’s a specific variant of the broader Azure BYOK GPT-5 issue we’re working on, distinct from the main one, so I’ve pinned your exact case with that request ID and passed it to the team. I don’t have an ETA on this particular one yet, but I’ll update here as it moves.
And no worries on the workarounds, glad you’ve got fallbacks covered in the meantime.
nice, thanks!
We’ve been merging some fixes here – could you let us know if you still face an issue?