Someone keeps an extended context window switched on permanently, treating it as a safety margin for every task they run. What is the cost of that habit?

The principle — Context is not free capacity. Cursor notes that a larger context window can increase the tokens used by a request, and that the default window works well for most coding tasks.

Why the key is correct — Usage is counted per request, so raising the ceiling raises what an ordinary request can consume. Left on permanently that applies to every small edit and quick question, so an allowance drains faster with nothing gained.

Why the others are wrong — Unused capacity is not free like spare disk space, so waiting until the window fills is the wrong model. Slower replies are not the documented cost; tokens are. And a bigger window does not pay for itself on work that did not need it.

Remember this — Turn the larger window on for the task that needs it, then off again.

Sources — Cursor's Max Mode documentation.