Cursor usage runs out or costs more than before
Recent tracked changes that can explain it:
What’s going on
The reason identical work can now generate a bill is that Cursor's plan stopped being a request count and became a spending budget. In June 2025 the old Pro allowance of 500 fast requests a month was replaced with $20 of model usage charged at API pricing, with anything beyond that billed as overage. The wrinkle that catches most people came in September 2025, when Auto mode stopped being unlimited for individuals and began counting toward that included usage — so if Auto was your default, requests you assumed were free are now drawing down the budget. Steady coding volume paired with a brand-new spend graph is the giveaway that the meter changed under you, not your behaviour. Cursor conceded the confusion itself: the CEO apologised and issued refunds for surprise overages billed between mid-June and early July 2025, so unexpected charges in that window were a known problem. Control starts in the usage dashboard — see which requests are actually metered and what each model costs per call, since frontier models can turn a handful of requests into a real charge. Set a hard spend limit so overage can't accrue silently, and keep the expensive models for work that needs them while cheaper or Auto-routed calls handle routine edits. If your monthly spend stays high, weigh the tiers: Pro+ at $60 gives roughly three times Pro's usage and Ultra at $200 about twenty times, either of which can beat paying overage on top of Pro. Usage-based billing rewards catching a spike early, so a periodic dashboard glance beats a month-end surprise.
Cursor has become more restrictive over time, mainly in pricing.
Some recent constraint changes — monitor before depending on it.
Related problems