On this page

Read the model breakdown

Cursor's dashboard shows requests, token breakdowns, and model usage. That is the first place to investigate a sudden jump. If you use a team plan, the admin dashboard adds member and project-level views that help separate normal work from an outlier.

Do not stop at the total. Look for a change in model mix, a shift into longer contexts, or background agents that continued working after the foreground task looked quiet.

Track the mode with the request

Auto, a named model, and Max Mode do not consume usage in the same way. Cursor notes that Max Mode enables larger contexts and longer reasoning, so it can use more allowance. Labeling a task only as “Cursor” hides the most important variable.

A useful personal log needs just four fields: task, model or mode, start percentage, and end percentage. Add outcome quality if you want to compare value, not just cost.

Keep your usage visible while you work. Get Super Notchy for $39 ↗

Keep the limit visible

Cursor provides dashboard and editor notifications. A separate at-a-glance tracker is useful when Cursor is one of several tools in the day. Super Notchy can show the current Cursor reading beside Claude Code and Codex, so switching models does not also mean switching usage pages.

The practical goal is to know whether the next agent run fits. A current percentage, the billing reset, and the model used on the last large task usually answer that question.

Sources and further reading