Reading the dashboard properly
The usage view shows two pools side by side, and the mistake is reading them as one bar. Cursor Models covers Grok and Composer and is the pool Cursor describes as carrying significantly more included usage. Other Models covers everything third-party and bills at each model's API price.
A full Cursor Models bar next to an empty Other Models bar is not a bug. It is the signature of someone who pinned a frontier model in week one. The fix is a setting change, not an upgrade.
Crossing the line is not a cliff
When included usage runs out, two options exist and neither stops your work. On-demand usage continues at the same API rates on pay-as-you-go billing, and Cursor is explicit that requests are never downgraded in quality or speed when you cross over. Upgrading a tier is the alternative, and is often cheaper than the on-demand line it replaces.
That makes the dashboard a budgeting tool rather than an alarm. The number worth watching is not whether you will run out, but what your on-demand line will look like by the end of the cycle.
Where the money actually goes
Three things dominate. The model you selected, how long your conversations run, and how many agents you have going at once.
The first is obvious and still the most common cause. The second is quieter: a long conversation resends its accumulated context on every turn, so the tenth message in a session costs more than the first even though it looks the same. The third catches people who discover worktrees and run four agents in parallel, which spends four allowances in the time one used to take.
The cycle date is yours, not the calendar's
Both pools reset with your monthly billing cycle. If you subscribed on the 14th, your allowance refills on the 14th. This matters when you are rationing near the end of a period, because the reset is probably not where you assumed it was.
Usage behaviour read from the models and pricing documentation on 25 September 2026.