The “require custom models” warning means that some Cursor features (Tab completions, Apply from Chat, and Agent) use Cursor’s fine-tuned models, which aren’t available on OpenAI’s platform. These features will always run through your Cursor subscription, not your API key.
Here’s how billing splits once you enable your OpenAI key:
Billed to your OpenAI Enterprise account - Chat and Composer requests where you manually select an OpenAI model (GPT-4.1, o3, etc.)
Billed through your Cursor (team) subscription - Tab completions, Apply from Chat, Agent (when it uses Cursor’s custom models for orchestration), Auto mode (routes through Cursor regardless of BYOK), Any non-OpenAI model usage (Claude, etc.)
All features continue to work normally. The warning is informational only —it lets you know that some features can’t be charged to an external key because they rely on models available only through Cursor.
For more details: API Keys / BYOK
A related thread with a detailed staff explanation: Need clarification with my own API key