I’m using a custom API endpoint in Cursor with a custom Base URL and a custom model name.
My current configuration is approximately:
Custom Base URL: enabled
Custom API Key: enabled
Model name:cx/gpt-5.6-sol-max
However, when I try to use this model, Cursor shows the following error:
This model does not support custom API keys.
This seems incorrect because cx/gpt-5.6-sol-max is a custom model served through my own OpenAI-compatible API endpoint. The API key and Base URL are both provided by me, rather than using Cursor’s built-in model provider.
Steps to Reproduce
Open Cursor Settings.
Configure a custom OpenAI-compatible Base URL.
Configure a custom API key.
Add a custom model with the name: cx/gpt-5.6-sol-max
Select the model in Cursor.
Try to send a request.
Cursor displays: This model does not support custom API keys.
Expected Behavior
When a custom Base URL is configured, Cursor should send the configured model name directly to the custom OpenAI-compatible endpoint using the provided custom API key.
Cursor should not reject the model based on whether the model name matches one of Cursor’s built-in models.
In other words, a model such as:
cx/gpt-5.6-sol-max
should be treated as an arbitrary model identifier belonging to the custom API endpoint.
Hey, thanks for the detailed report. This isn’t anything in your setup. We’re tracking this issue. It happens when a custom model name contains the name of a built-in Cursor model inside it, in your case gpt-5.6-sol inside cx/gpt-5.6-sol-max. Because of that, the request incorrectly hits the guard “This model does not support custom API keys”, even though the key and Base URL are yours.
Workaround, if your gateway supports it: give the custom model a name that doesn’t overlap with built-in model names, so avoid substrings like gpt-5.6-sol, gpt-5.6-terra, etc. From your config, your other custom models with neutral names like glm-cn/..., oc/... work through the same endpoint without issues, so renaming or using an alias usually fixes it.
We’ll ship a proper fix on our side, but I can’t share an ETA yet. I’ll post in the thread when there’s an update. If renaming doesn’t work for you or doesn’t help, let me know.
Hey, I promised to come back with an update, here it is.
We fixed the incorrect error message “This model does not support custom API keys” for custom models with names like cx/gpt-5.6-sol-max. The change is rolling out on our side and may already be live.
One important note: if a custom model’s name includes the name of a built-in model, that custom model follows the regional availability rules for that built-in model. Model availability can vary by region under our Terms of Service and Acceptable Use Policy, and we enforce these rules on an ongoing basis. So in regions where that model isn’t available, instead of the API key message you’ll now see a more accurate message about regional availability.