Cursor Usage Limits: Rate Limits and Included Usage

Cursor tracks two monthly pools instead of one request count. The Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool currently covers Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5, and Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition.. The Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool covers third-party models at their API prices. The table shows how those pools differ across the current paid individual plans.

The Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool currently includes Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5, and Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition.. The Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool holds third-party models charged at each model's API price. Both balances appear in editor settings and on the usage dashboard. The current model guide keeps the selection side separate from this budget page.

Which model you pick decides how fast that budget drains, because each model bills at its own API rate. Cursor's price table puts Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. at $0.50 for a million input tokens and $2.50 for a million out. GPT-5.6 Sol is $5 and $30 in those same two columns, which is a tenfold spread between two choices a sensible person could make. Cache reads are cheaper again, at $0.20 per million on Composer 2.5.

Max Mode now belongs only to legacy request-based plans. It extends a supported model beyond the default context limit and is billed at that model's API rate plus 20%. Do not plan a current usage-pool budget around a Max Mode toggle.

Tab completions sit outside these model pools. Cursor lists unlimited tab completions on Pro, Pro Plus, and Ultra, so the dollar figures above describe model usage rather than every editor action. The Cursor StartAn India-only individual plan centered on Cursor Models, with cloud and mobile access but without Other Models, Auto, Bugbot, Automations or the Agent SDK. Press Enter for the full definition. guide covers that plan's narrower model and product scope.

The full picture lives in the web dashboard. Go to cursor.com/dashboard and open the Usage section for real-time usage, remaining allowance and any on-demand charges. Both pool balances also show in your editor settings, and that is the faster check mid-task.

1. 1Open cursor.com/dashboard and find the Usage section.
2. 2Read your real-time usage, remaining allowance and any on-demand charges.
3. 3Review the token breakdown to see which models are consuming the budget.

The token breakdown in step three is the one to actually read. A remaining balance tells you how much is left and nothing else. The breakdown tells you which model spent it, and two thirds gone on the 9th reads very differently once you can see that one long-running task on an expensive model accounts for most of it.

How often to look is a question about how you are paying, more than about how much you use. With on-demand off, the limit is its own alarm and a weekly check is reassurance. With it on, the dashboard is where an on-demand charge shows up before the invoice does.

Start with the token breakdown on the usage dashboard, since it splits spend by model. More often than not the answer is that one model did most of it.

An expensive model left selected in the picker is the usual culprit. Nobody picks GPT-5.6 Sol to rename a variable, but a model chosen on Friday for a hard problem is still the model chosen on Monday, and the price spread between Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. and GPT-5.6 Sol applies to every routine edit made in between.

Context size does the same thing more quietly. A larger window means more tokens read on every turn, so an identical prompt draws more on a wide window than on a narrow one when the task fills that extra space. Max Mode explains that legacy request-plan option and what the additional context buys you.

The cache is the least visible of the three. Cached input is priced below fresh input, $0.20 against $0.50 per million on Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., and switching models partway through a thread breaks the cache, so a long conversation that changed models three times has been paying full price for context it had already sent.

One correction to where I started. The token breakdown is the right tool for a bill that came in larger than expected, but if the complaint is that you ran out, check which pool emptied before you open the breakdown at all. The two drain independently, so running out is usually a statement about the Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool and not about the plan as a whole. Both balances sit in your editor settings.

Cursor does not silently switch a request to lower quality or speed. Once included monthly usage is exhausted, you choose whether to add on-demand usage at the same published rates or move to a higher plan. Start does not include on-demand usage, so its path is an upgrade.

• Enable on-demand usage on an eligible plan so work continues at the same API rates.
• Upgrade to a higher plan, which includes a larger monthly usage budget.

Which of the two is cheaper is arithmetic, at least on the individual plans. On-demand bills at the same API rates as your included usage, so an extra $50 of work costs $50. Moving from Pro to Pro Plus costs $40 more a month and adds $50 of included API usage, so past $40 of overage in a month the upgrade is cheaper, while one unusual month is not. That arithmetic only covers Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. because Cursor publishes no dollar figure for its generous Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. allowance.

The part worth planning for is the timing rather than the money. The notification arrives when it arrives, and that tends to be partway through an agent run on a Thursday afternoon, with a billing decision attached that nobody wanted to make right then. Deciding in advance costs nothing. Pick on-demand or a plan tier while you are not blocked, and set it once.

Usage resets monthly, keyed to when you subscribed rather than to the first of the month. The details below matter most to teams reconciling a budget across a group of seats.

The anchor is the row that catches people out. Subscribe on the 15th and your usage month runs 15th to 15th, while the finance calendar almost certainly runs 1st to 1st, so the two disagree by half a cycle every time anyone sets them side by side. On one seat that is a curiosity. It becomes a real problem when somebody is reconciling a Cursor line against a monthly budget and cannot see why the totals never land.

There is no rollover, so saving budget for next month is not a strategy. Whatever you have not spent by your reset date disappears. Check that date under Manage Subscription before you plan around it.

On a team the personal anniversary stops mattering, because everyone resets together on the team billing cycle.

Start by moving suitable work to the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool you already have. If that does not fit the task or the pool is exhausted, the remaining choices are billing changes.

• Use a Cursor Model directly when Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5, or Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. fits the task. That work draws from the separate Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool.
• Enable on-demand usage to pay for additional requests at the same API rates.
• Upgrade your plan. Higher tiers include more usage.

The first item uses a pool already included with the plan, which is why the order is worth taking literally. Cursor describes its allowance as generous but does not publish a dollar figure. That makes it useful for routine work and unsuitable as the only number in a finance forecast, so keep the Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. amounts from the first table in the budget.

On a team plan there is a gap in that second option, and it is worth knowing before you lean on it. Spend alerts watch on-demand spend only, so a threshold cannot warn you that the included budget is about to run out. Enterprise accounts with pooled usage are the exception, where member-level alerts can be set on total spend. Alerts notify rather than stop, so spend limits are the control that actually caps, and the billing and account guide has the setup.