Cursor limits depend on the plan and the surface. Hobby has limited Agent and Tab usage but no published request count. Paid individual plans make Tab unlimited and split agent usage into generous access to Cursor models plus a dollar allowance for third-party models. Pro includes $20 of third-party model usage, Pro Plus $70, and Ultra $400 in the August 2026 site snapshot. Those allowances refresh monthly on the account billing cycle. When the third-party pool is empty, a user can wait, switch pools, upgrade, or enable on-demand usage at model API rates. Team admins can add spend controls on top.
- Cursor has several independent limits. Identify the empty pool before changing a plan or model.
- Hobby numbers are not published. The account dashboard is authoritative for a free account.
- Tab is unlimited on paid plans. Agent model usage is the part that normally reaches a ceiling.
- Paid plans split usage into Cursor Models and Other Models, and the two balances do not drain together.
- Included usage refreshes on the monthly billing cycle, even when the subscription itself was paid annually.
- On-demand usage is optional continuation at model rates, not a slower fallback queue.
The limit map in one table
Start with the exact thing that stopped. A missing Tab suggestion, a blocked Claude request, an Agent message that routes to Composer, and a team spend warning can all be described as "the Cursor limit", but they come from different meters. Reinstalling the editor cannot change any of them because the enforceable balance lives on the account, not in the local application.
Plan and usage structure reflected by the site pricing snapshot, checked August 2026.
| Limit | Who has it | What reaches the ceiling | What happens next |
|---|---|---|---|
| Hobby Agent | Free accounts | Agent and Chat activity on the available model route | Agent requests stop until the account cycle refreshes or the user upgrades |
| Hobby Tab | Free accounts | Tab requests, including suggestions that are requested but not accepted | Suggestions stop until the account cycle refreshes or the user upgrades |
| Cursor Models pool | Paid individual and team seats | Composer, Grok, and other models identified by Cursor as first party | Use another eligible pool, wait for refresh, or move to a plan with more capacity |
| Other Models pool | Paid plans | Claude, GPT, Gemini, and other third-party models, priced by tokens | Enable on demand, switch to a Cursor model, wait, or upgrade |
| On-demand spend limit | Accounts with overage enabled | Usage after the included third-party allowance | Requests stop when the configured hard spend ceiling is reached |
| Team or organization cap | Managed accounts | Per-user or organization spend according to admin policy | The admin policy wins even when the underlying plan still has capacity |
| Provider or service capacity | Any account | Temporary model or Cursor service availability | Retry, change model, or check status; this is not an account quota |
Do not translate the current system back into "requests per month". A request that asks a cheap model to answer a narrow question can cost a fraction of a long Agent run that reads a repository, calls tools, and emits a large diff. That variability is precisely why Cursor moved its paid individual plans from request counting to token-priced model usage. The old number was easy to repeat and poor at predicting real capacity.
Cursor free usage limits
The free tier is called Hobby. It requires no credit card and does not expire, so it is a genuine tier rather than a fourteen-day countdown. It includes the core editing experience, limited Tab completions, and limited Agent requests. Cursor does not publish a fixed current request count for Hobby. Any page promising exactly 2,000 completions, 200 small-model uses, or 50 slow premium requests is quoting an older plan generation.
That absence of a public number changes how to use the free plan. Hobby is suitable for deciding whether Cursor's editing loop fits: whether Tab predicts useful edits, whether Inline Edit feels natural, and whether you like reviewing Agent diffs in the editor. It is not suitable for capacity planning. The only useful allowance is the one shown on the dashboard for the signed-in account.
| Hobby surface | Limit shape | Practical consequence |
|---|---|---|
| Tab | Limited monthly quota; exact public count omitted | Automatic requests can consume allowance even when a displayed suggestion is not accepted |
| Agent and Chat | Limited access on the available route | Long, tool-heavy jobs reach the ceiling sooner than short questions |
| Third-party model allowance | None as a separate dollar pool | Do not budget Hobby around pinned Claude or GPT usage |
| Cloud agents and paid extensions | Paid-plan capability | A free usage ceiling is not the only reason a control may be unavailable |
| Expiry | None | Evaluation can happen over real work instead of during an artificial trial sprint |
Free Tab usage can look especially strange. Cursor requests predictions as you pause and type; the counter is not simply the number of completions you accepted. A developer can therefore see a modest accepted-completion count and still reach the free ceiling. That is a meter-definition issue, not proof that the account is corrupt. On a paid individual plan, Tab is unlimited, which removes this particular failure mode.
Cursor Pro usage limits and the paid ladder
On paid plans the feature set is broadly similar and the ladder mainly sells more agent capacity. Pro is the standard individual tier at $20 a month. The site snapshot describes Pro Plus as three times Pro Agent limits and Ultra as twenty times Pro Agent limits; it does not expose stable public sticker prices for those higher individual tiers, so the signed-in plan picker is the correct source for upgrade arithmetic.
Included third-party model allowances in the August 2026 site pricing reference.
| Plan | Tab | Cursor Models | Other Models allowance | Capacity label |
|---|---|---|---|---|
| Hobby | Limited | Limited | No separate allowance | Evaluation |
| Pro | Unlimited | Generous included usage | $20 monthly | Daily editing, limited Agent work |
| Pro Plus | Unlimited | Generous included usage | $70 monthly | 3x Pro Agent limits |
| Ultra | Unlimited | Generous included usage | $400 monthly | 20x Pro Agent limits |
| Start, India only | Paid-plan access | Generous included usage | $0 | Cursor models without a third-party pool |
The dollar allowance is not cash and does not roll forward. It is the amount of third-party inference the plan includes for that monthly cycle, evaluated at the selected model's rate. A $20 allowance can represent many short, cheap requests or a much smaller number of long frontier-model runs. Model choice, context size, generated output, tool results, and the number of Agent turns all change the drain.
The advertised multipliers are also not a promise that every meter literally displays three or twenty times one number. Cursor can grant bonus capacity, change first-party inclusions, and route models differently. Read the multiplier as the product tier position; read the account dashboard as the enforceable capacity. This distinction matters when someone compares their own total with a screenshot from another account and assumes one is wrong.
The two model pools
The current model is easier once you stop picturing a single battery. Paid Cursor has two batteries. The first covers Cursor Models, including Composer and the Cursor-branded Grok route named in the current site content. The second is a dollar-denominated allowance for Other Models supplied by companies such as Anthropic, OpenAI, and Google. Using one does not necessarily move the other.
| Choice | Pool | How to reason about it |
|---|---|---|
| Tab completion | Neither paid Agent pool | Unlimited on Pro, Pro Plus, and Ultra |
| Composer or another listed Cursor model | Cursor Models | Use for routine Agent work when preserving third-party allowance matters |
| A manually pinned Claude, GPT, or Gemini model | Other Models | Token usage is valued at that model rate against the plan allowance |
| Auto Cost | Other Models under the current router description | Optimizes for price and uses a flatter, more predictable cost treatment |
| Auto Balance or Intelligence | Other Models when routed to them | Billed according to the model selected by the router |
| Legacy Max mode | Legacy pricing behavior | Do not apply an old Max explanation to a current token-based plan without checking account generation |
This is why two users on Pro can report completely different experiences without either being mistaken. One spends the month on Tab and Composer and barely touches the third-party allowance. The other pins a frontier model, opens a large monorepo, and runs several long agents a day. The subscription label is identical; the expensive pool usage is not.
It is also why the correct response to a low balance is often routing rather than rationing. If the task is mechanical, move it to Tab or a Cursor model. If it is an architecture decision where the chosen frontier model is the point, spend deliberately. Saving an expensive model for work that benefits from it is a better policy than treating every prompt as equally valuable.
Cursor on-demand usage
On-demand usage is the bridge past the included third-party allowance. Once enabled, eligible requests continue at the same model API rates and the additional amount is billed in arrears. Cursor says requests are not downgraded in quality or speed when they cross from included usage into on demand. That makes the failure mode financial rather than technical: work continues, but the invoice can exceed the subscription price.
Decide whether continuity is worth variable cost
A deadline-driven individual may prefer an explicit overage ceiling. A student or fixed-budget team may prefer a hard stop. Neither setting is universally safer; it depends on whether money or interruption is the unacceptable surprise.
Set a spend limit before the included pool is empty
Do not wait for the error and then enable an unbounded continuation reflexively. Pick a maximum that represents a conscious purchase, not a blank authorization for the rest of the billing cycle.
Compare repeated overage with the next tier
One exceptional week usually favors on demand. Similar overage in two consecutive cycles is evidence that the plan no longer matches the workflow. Compare the actual overage with the current signed-in upgrade price.
Keep first-party routing available
A hard cap should not turn an empty third-party pool into a total work stoppage. Know which Cursor model is acceptable for routine tasks before the cap is reached.
On demand does not increase the included allowance and does not make the next reset arrive sooner. It creates a second billable region after the allowance. If a hard spend limit is reached, the account can still stop despite on demand being enabled. That is expected: the spend control is doing exactly what it was configured to do.
When every limit resets
Cursor usage is monthly, but "monthly" does not mean midnight on the first day of the calendar month. The billing cycle starts from the account's subscription date. The dashboard and Manage Subscription view are the authority for the next renewal or reset date. An annual plan still meters included usage in monthly slices.
| Meter | Reset behavior | Where to verify |
|---|---|---|
| Hobby limited usage | Monthly account cycle; public numeric allowance omitted | Cursor dashboard for the signed-in account |
| Pro / Pro Plus / Ultra included usage | Monthly billing cycle | Dashboard usage view and Manage Subscription |
| Annual subscription included usage | Still refreshes monthly | Billing portal; annual refers to payment commitment |
| On-demand total | Accumulates within the billing period and is billed in arrears | Usage and billing views |
| Team per-user or organization cap | According to the current team billing cycle and admin policy | Admin dashboard |
| Temporary service rate limit | Operational, not subscription-calendar based | Cursor status and the request error |
Upgrading can start a new subscription cycle and replace the prior allowance, but it should not be used as a reset button without reading the checkout terms. A new tier may be prorated or start a new cadence; prior on-demand spend does not vanish merely because the included pool changes. The account checkout is the source of truth for that transaction.
The practical habit is to record the reset date beside the remaining balance. A balance without time is not a capacity forecast. Twenty percent remaining with two days left is healthy; twenty percent with eighteen days left is a routing or plan problem. This is the point at which tracking becomes useful, but the tracking page should own the dashboards and methodology rather than this mechanics reference.
Team limits are policy on top of capacity
A managed Cursor account can stop before the vendor-level plan pool appears empty because the organization adds controls above it. Teams can use seat types, per-user rules, central billing, and spend limits. Enterprise arrangements can pool usage and negotiate terms. A developer looking only at a personal usage widget may therefore see available capacity while an organization policy denies the next request.
The June 2026 Teams structure described in the site content gives Standard seats separate first-party and third-party pools and introduces a Premium seat at five times Standard Agent limits. Standard remains the published $40 monthly seat in the site snapshot. The reason to buy the team tier is administration, policy, and one invoice, not a claim that every developer gets a cheaper version of individual Pro.
- Developers: capture the exact error and selected model before asking an admin for more budget.
- Admins: distinguish an included-pool warning from a hard organization cap in the message you send.
- Finance: review on-demand and subscription charges separately; they answer different questions.
- Engineering leads: do not give every user the power-user seat by default. Measure who repeatedly reaches Standard first.
- Everyone: use the account and admin dashboards from the same billing cycle when reconciling a discrepancy.
What to do at each ceiling
| What stopped | First action | Do not do first |
|---|---|---|
| Hobby Tab | Check the dashboard reset date; upgrade only if daily Tab is now essential | Reinstall Cursor |
| Hobby Agent | Wait, upgrade, or move the task to another tool you already fund | Search for a retired free request count |
| Third-party model on paid plan | Switch to a suitable Cursor model or decide on capped on demand | Assume every Cursor feature is exhausted |
| Cursor Models pool | Check the plan-specific balance and alternative eligible routes | Buy third-party overage without knowing which pool is empty |
| On-demand hard cap | Raise it intentionally, upgrade, or wait for the cycle | Disable budget controls merely to clear a toast |
| Team policy cap | Ask the admin with the exact model, time, and error | Create another personal account to bypass policy |
| All users or all models slow/failing | Check status and network diagnostics | Treat an operational incident as a subscription limit |
The order is deliberate: diagnose, route, then buy. An upgrade can be correct, but it is the expensive answer to a question that might actually be "why was a frontier model pinned for routine work?" Conversely, endless optimization can cost more in attention than the next tier. If the same well-scoped professional workload exhausts the same pool twice, capacity is the problem and buying capacity is rational.
Questions people ask
There is no single limit. Hobby has limited Agent and Tab usage with no current public numeric count. Paid plans have unlimited Tab plus separate pools for Cursor models and third-party models. On-demand and team spend caps can add further ceilings.
Pro includes unlimited Tab, generous usage of Cursor models, and $20 of monthly third-party model usage in the August 2026 site snapshot. The number of requests varies because models and prompts consume different token amounts.
Cursor describes Hobby as limited Agent requests and limited Tab completions, but does not publish a fixed current request count. Check the signed-in dashboard for the allowance and reset date governing your account.
The subscription allowance is monthly on the account billing cycle, not a daily reset. Temporary operational rate limits can exist, but they are different from the plan allowance.
Included usage refreshes on the monthly billing cycle shown in the Cursor dashboard or Manage Subscription view. Annual subscribers still receive monthly usage refreshes.
Optional paid usage after the included third-party model allowance is consumed. It is billed in arrears at model rates, without a quality or speed downgrade. Set a deliberate hard spend ceiling.
Model choice and Agent shape determine token cost. A pinned frontier model, long context, large outputs, and many tool turns can consume the third-party allowance far faster than Tab plus first-party Cursor models.
No. Annual billing changes how the subscription is paid, while included model usage is still accounted for and refreshed in monthly cycles.
First identify the empty pool and check for an accidentally pinned expensive model. If the same necessary workload exhausts the same allowance in two cycles, compare the next plan price with actual on-demand spend.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.