Keep Cursor Auto Inside Your $20 Plan

You open Cursor on a Saturday morning and ask the agent to clean up a small module. By lunch, the usage total has moved, and the old comfort of one predictable number is gone. Cursor ended its single flat Auto pricing on August 24, 2026. On August 24, 2026, subscriber dashboards displayed $500 per month for third-party models and $3,000 per month for Cursor models, after the Cursor-model amount was increased from $2,000 that day. For a weekend builder, the change turns Cursor Auto from a flat habit into a metered tool: you choose the model, you watch the pool, and you decide when to stop.

The flat number is gone

The retired model-independent Auto price had four parts: $1.25 per million input tokens, $1.25 per million cache-write tokens, $0.25 per million cache-read tokens, and $6 per million output tokens. Cursor's customer email said the new Auto pricing would mean a higher rate for most requests. The public plan table still labels both monthly usage pools as Included for Pro at $20, Pro+ at $60, and Ultra at $200, but it does not state their dollar values. When the dashboard shows the two pools, treat them as the usage list you can actually read. The public page tells you what is Included; the dashboard tells you the number that matters before the next decision.

Pin the cheap model before the agent starts

Auto mode is useful because it picks a model for you. It is also useful because it can pick the wrong one for your budget. For routine edits, small refactors, test fixes, and documentation, pin a cheaper model in the model picker and save that choice for the next time you open the same kind of task. The agent should not decide cost for you unless you want it to.

The dashboard is the clearest check. It does not tell you whether the code is good. It tells you whether the weekend is still affordable. Agent work is hard to price by the hour, because the agent can run for minutes or hours depending on the task. Before a long run, look at the two pools and decide if the task is worth the next amount of usage. If the answer is no, make the task smaller or move it to a later run.

Pin, watch, and cancel before the weekend ends

Use this list before a long agent run. Each item is a check you can do before the next task, not a monthly ritual.

• Pin a low-cost model for routine work. Open the model picker and choose the model you want for small edits, tests, and docs. If the task is exploratory, leave Auto for that one run only.
• Set a personal stop point. Write a number you will treat as a limit, such as a weekly usage target or a dollar amount you are willing to spend beyond your plan. The number is yours, not Cursor's.
• Check the dashboard before a long run. Look at the dashboard before you ask the agent to refactor, migrate, or debug across many files. If the pool is already close to your limit, shrink the task.
• Stop the agent when the limit is reached. When the dashboard crosses your personal stop point, pause the run and finish the remaining work in smaller steps. Do not let a single agent session decide the whole weekend.
• Cancel or downgrade before renewal. If you cross the limit twice in a row, change the plan before the next billing date. A cheaper plan is a normal outcome.

The same check covers team and enterprise plans

For Teams and Enterprise, third-party Router requests add a Cursor Token Rate of $0.25 per million tokens. Enterprise Auto Cost pricing remains fixed at $1.25 per million input tokens, $1.25 per million cache-write tokens, $0.25 per million cache-read tokens, and $6 per million output tokens until September 7, 2026. If you are solo, those lines are less relevant, but they still show the same habit: check the dashboard before the next run. Keep the tool inside the hours you actually have: a short run, a pinned model, a visible pool, and a stop point you set yourself before the next task.