Halfway through a billing cycle, a team running Auto in Intelligence mode finds its included usage nearly exhausted although request volume matches previous months. What explains this, and what can be adjusted?

The principle — The optimisation mode changes what a request costs, not merely how good the answer is. The higher modes route to more capable models and bill at those models' rates, so they draw down the usage limit faster than the cost-optimised mode does. The mode is a setting, and it can be switched at any time.

Why the key is correct — With request volume unchanged, the variable that moved is the price of each request. A team sitting on a quality-optimised mode spends the same included allowance across fewer requests, which looks from the dashboard exactly like the plan having shrunk. The remedy is immediate rather than administrative: drop the mode for routine work, keep the higher setting for the tasks that genuinely need it, and the burn rate falls from the next request onward. Nothing waits for the cycle to turn over and no plan change is involved.

Why the others are wrong — Treating the mode as a quality-and-latency dial is the belief that hides the cause, sending the team to audit their subscription rather than their model picker. The mode is also not frozen for the cycle; it sits under Auto in the picker and applies to the next request. Throttling is not part of the design either: when included usage is exhausted, requests either stop or begin billing as on-demand charges, according to whether on-demand usage is enabled.

Remember this — The optimisation mode is a spending control. Higher modes buy quality with budget, and you can turn it down mid-cycle.

Sources — Cursor's Cursor Router documentation.