Custom OpenAI override and Cursor models cannot both stay enabled

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Cursor version: 3.21.18

OS: macOS 26.5.1

I have a local model exposed through an OpenAI-compatible endpoint. In Settings → Models I turn on the OpenAI API key and Override OpenAI Base URL so custom models go to that endpoint.

With the override on, Cursor’s own models stop working. Requests for those models are treated as if they use the custom key and endpoint, and they fail (for example “This model does not support custom API keys” or “Invalid API key. Unauthorized User API key”). Request ID: 455683a2-9e6d-483d-a7b7-7c889c14dee2.

With the OpenAI key and override turned off, Cursor models work again, and the custom models no longer reach the local endpoint.

Expected: both can stay on at the same time. Custom models use the OpenAI key and override base URL. Cursor’s built-in models stay on Cursor inference. I should not have to toggle the override off to use a Cursor model and back on to use the local one.

Actual: the override is global. Enabling it breaks Cursor models. Disabling it breaks the local custom route.

Steps to Reproduce

Run an OpenAI-compatible server locally.
In Settings, Models, turn on the OpenAI API key and Override OpenAI Base URL, and point the URL at that server.
Add a custom model and send a message. It should reach the local server.
In the same window, switch to a built-in Cursor model and send a message.
The Cursor model fails (invalid API key, or “this model does not support custom API keys”).
Turn the OpenAI key and override off. The Cursor model works, and the custom model no longer reaches the local server.

Expected Behavior

Both stay on at once. Custom models use the OpenAI key and the override base URL. Built-in Cursor models stay on Cursor’s own inference. No toggling between them.

Operating System

MacOS

Version Information

Version: 3.21.18
VSCode Version: 1.128.0
Commit: c4730f7d93d787d9ab120af715999f0345ee5bc0
Date: 2026-09-22T02:04:34.641Z
Layout: glass
Build Type: Stable
Release Track: Default
OS: Darwin arm64 26.5.1

For AI issues: which model did you use?

A built-in Cursor model (for example Grok) with the OpenAI override enabled, and a custom model pointed at the local OpenAI-compatible server.

Does this stop you from using Cursor

Yes - Cursor is unusable

7

Hey @chrono, thanks for the report.

The OpenAI API key and Override OpenAI Base URL apply to every model that is not a Claude or Gemini model. That includes Composer and Grok. Those models run on Cursor’s infrastructure, so with your key attached they are rejected with “This model does not support custom API keys”. Keeping your local models on the override and those Cursor models on Cursor at the same time is not available today. The team is aware of the ask here.

Until then:

  • Claude and Gemini models are not affected by the OpenAI key. They only use their own Anthropic or Google key toggles.
  • For Composer and Grok, turn the OpenAI key off in Cursor Settings > Models.