Use your own API key in Cursor, and what it costs

Cursor lets you paste an OpenAI, Anthropic, or other API key and use it for chat. A few rules decide what that really covers:

• Your own key is used for chat models only. Tab completions stay on Cursor's own models, whatever key you add.
• Your own API keys need a Cursor plan that allows them. Cursor Plus does not; Premium and above do. Check Cursor's pricing page for your plan.
• On Cursor Teams and Enterprise plans, Cursor adds its Cursor Token Rate to eligible third party models, even on your own key. Its docs list $0.25 per million tokens (input, output, and cached) as of Oct 27, 2026. Individual plans don't pay it. Check Cursor's pricing page for today's terms.
• While Override OpenAI Base URL is on, Cursor sends every OpenAI model through that address. Turn it off to use Cursor's own OpenAI models again.

Three bills, at most. Your Cursor plan stays the same. Your API provider bills your key for the tokens you use, at its list price. On a Cursor team plan, Cursor's own fee may apply on top.

If you point Cursor at FUSR.ai, that does not change: FUSR.ai adds nothing to token prices, and Anthropic, OpenAI, and xAI still bill your keys directly. What changes is which model answers. With fusr/auto, a quick question can go to a cheaper model while a hard one stays on a strong one, prompt caching keeps working, and the cap you set stops a runaway session.

To get a feel for the token part, try the AI API cost calculator.

First connect at least one API key on FUSR.ai and get your FUSR.ai key. The quickest way is npx fusrai setup in a terminal: it signs you in, gets your key, and prints these same steps. Cursor keeps its settings in its own app storage, so the command never edits Cursor itself. Add --copy and it puts the key on your clipboard for you.

Then, in Cursor:

Cursor sends these requests from its own servers, which reach the proxy over the internet. If Verify fails, check the base URL has no trailing slash and that your FUSR.ai key is the one shown on your Routing page in FUSR.ai.

• Keep one chat on one topic. A chat that stays on one model keeps its prompt cache warm.
• Name a model when you want it: fusr/claude, fusr/openai, or fusr/grok. fusr/auto picks for you.
• Set a spend cap on your FUSR.ai key: a monthly cap on any plan, or a daily one on Pro and Team. An agent that loops stops at the cap instead of at the end of the month. More ways to cut the bill.