Struggling with Azure OpenAI setting

Support Help
21

Has this been fixed? It currently only works with 4o

22

Do reasoning and max mode even work for BYOK at all ?
@danperks is there anything in the roadmap for this?

  • not having parity between self hosted and BYOK while having something like max mode with it’s per api-call pricing seems like an inconsistency in the product positioning by cursor, if reasoning and max mode are not going to be offered for BYOK why even have BYOK at this point ?
23

The difficulty with API key support in general is various models and providers all have different formats for how they need requests to be structured.

I’ve bumped this feedback up again internally to see if it’s something we can resolve in the future, but due to the 1000s of things we want to work on and the engineering capacity we have, BOYK is currently offered as a “best effort” feature until we can dedicate some time to give it some love or a revamp.

24

@ksquarekumar I’ve created this project for the community, which allows you to use Azure GPT-5 in Cursor GitHub - gabrii/Cursor-Azure-GPT-5: A service that allows Cursor to use Azure GPT-5 deployments.

It’s not perfect, but useful until model support is improved by the team. And after building this, I can attest to the complexity of supporting different BYOK configurations

1 Like
25

Question, the app currently asks for endpoint url, deployment name, and api key. However, Azure needs api version as well.

Currently gpt-4o works but gpt-5 doesn’t work. I think this is most likely due to api-version changing. When I looked at cursor code there is no place that this api-version could be changed.

Could you update this on your end? this should be a quick fix.

26

Hi @Mehrdad , it’s not just because of the api version. It’s because cursor currently doesn’t support BYOK for models using the responses api (like GPT-5). That is why gpt-4o works (as it supports the completions api).

For example, my project works by allowing cursor to consume GPT-5 as if it was a custom model served through the completions api, while behind the scenes it’s using azure through the responses api.

1 Like