Cursor Pricing in 2026: Plans, Usage Limits, and What the Public Pages Do Not Quantify

Last updated: August 15, 2026

Cursor meters AI work through two separate monthly pools on Pro, Pro Plus, Ultra, and Teams. India’s Start plan carries one. The documentation publishes a dollar allowance for third-party models on the three global individual tiers, plus per-token rates for the models it lists. It does not publish an absolute monthly allowance for the Cursor Models pool on any page reviewed here.

Verdict: Cursor publishes enough to price a known volume of tokens, and not enough to calculate what an individual plan includes before purchase.

Which plans have which pools

Cursor separates model usage into two pools that are metered and reset independently.

Cursor Models covers Cursor Grok 4.6, Cursor Grok 4.5, and Composer 2.5. Other Models covers third-party models and is charged at the provider’s API price. The documentation states both reset with the monthly billing cycle and that unused usage does not carry forward. On team plans, every member’s usage resets together on the team billing cycle.

Start, the plan for developers in India, is the exception. It carries the Cursor Models pool and does not include the Other Models pool, on-demand usage, Bugbot, Auto, Automations, or the Cursor SDK. Its three models run in non-fast mode only, and both Grok versions are fixed at medium effort.

The two multipliers on the pricing page carry a specific ambiguity. Pro Plus is described as three times Pro’s Agent limits while its published third-party allowance moves 3.5 times, from $20 to $70. Ultra’s stated twenty times matches its $400 allowance exactly. Neither page states which pool the multiplier measures.

Sources: Cursor Docs, Models & Pricing, read 15 Aug 2026 · Cursor, Pricing, read 15 Aug 2026

Four levels of disclosure

Cursor’s pricing is not opaque. It is unevenly disclosed, and the unevenness follows a pattern a buyer can work with once it is named.

The gap sits in one row. Cursor publishes the marginal price of a token in five places and the size of the Cursor Models pool in none of them. Pro Plus at three times and Ultra at twenty times are ratios against a base that is not stated, so they multiply an unknown.

Cursor’s own usage guidance is the closest public substitute. The documentation estimates that daily Tab users always stay within $20, limited Agent users often stay within the included $20, daily Agent users typically reach $60 to $100 in total monthly usage, and power users running multiple agents often exceed $200. Read against the plan table, that is the vendor stating that its $20 tier is sized for light Agent use.

Sources: Cursor Docs, Models & Pricing, read 15 Aug 2026 · Cursor, Pricing, read 15 Aug 2026 · Cursor, Pricing Policy, 21 May 2026

The published rates

Selected current rates, per million tokens. The documentation table continues behind a “Show more models” control, so this is a subset rather than the full list.

Fast raises Composer 2.5 input and output rates sixfold. Its cache read multiplier is 2.5 times, which is lower, so the effective increase depends on how much of a session is served from cache.

These rates also explain why a fixed request count would be a weak allowance unit. Consumption varies with the model, Fast mode, effort level, cache behavior, context size, and output length. Two developers sending the same number of requests can consume materially different amounts. That is a reasonable argument against publishing “500 requests” as an entitlement.

It is not an argument against publishing a denominator. Cursor already meters in tokens and publishes token rates. A dollar-equivalent, a token-equivalent, or a worked example would let a buyer compare tiers without implying that every request costs the same.

Sources: Cursor Docs, Models & Pricing, read 15 Aug 2026

What happens at the limit

The exhaustion path decides whether an underestimated allowance produces a stopped workflow or an invoice.

On Pro, Pro Plus, and Ultra, the help center states that the editor shows a notification when included usage is exhausted. The user can enable on-demand billing, which continues at the same API rates and is billed monthly in arrears, or upgrade the plan. The documentation adds that requests are never downgraded in quality or speed.

Start behaves differently, because it has no Other Models pool and no on-demand usage. What occurs when its Cursor Models pool is exhausted was not stated on the pages reviewed.

Two controls matter for team buyers. The Pricing Policy states that account administrators are granted authority to set hard limits or caps on organization usage so that cost ceilings can be monitored and managed directly. On Enterprise plans, Precommitted Usage is pooled across users under the same account number, which is not stated for Teams.

Sources: Cursor Docs, Models & Pricing, read 15 Aug 2026 · Cursor, Pricing Policy, 21 May 2026, sections 2.3 and 7.2

Auto, Router, and cost

Auto mode is powered by Cursor Router, which shipped on 22 July 2026. The documentation states that Cursor Router is currently available only on Teams and Enterprise plans.

Cost mode keeps Auto’s bundled rate regardless of which model runs. Balance and Intelligence are billed at the routed model’s own rate. Cursor’s wording is that these average about twice Cost, and rise to between two and four times depending on the mode selected.

Three operating conditions are documented.

Cursor Grok 4.5 must be enabled for the router to work. Blocking too many models degrades routing quality and can cause the router to be disabled, so a model allowlist policy and a routing policy interact.

The routed model is hidden by default, and the documentation states that hidden is both the default and the recommended setting, on the reasoning that results should be judged on their own merit rather than by model name. An administrator can switch the setting to displayed.

Administrators can enable the router per team or per organization group, restrict which optimization modes members may select, set the default mode, and enforce Auto softly or hard. On Enterprise the router is off by default and must be enabled.

Router requests may draw from either pool depending on which model handles the request, per Cursor’s help center.

Sources: Cursor Docs, Cursor Router, read 15 Aug 2026 · Cursor Changelog, 22 July 2026

The Cursor Token Rate and BYOK

On Teams and Enterprise plans, third-party model requests carry a Cursor Token Rate of $0.25 per million tokens on top of the model’s API pricing.

The documentation states that this rate applies to included usage, on-demand usage, and bring-your-own-key usage. Eligible BYOK requests therefore retain a Cursor token charge even though the inference is billed by the customer’s own provider account.

Auto Cost is exempt. All first-party Cursor models are exempt, and the help center adds that Composer 2.5 requests never incur the rate on any plan.

Sources: Cursor Docs, Models & Pricing, read 15 Aug 2026

Residency changes the price

Data residency is a paid option, and it belongs in a pricing review because it moves the rate table.

Opting in to regional data residency incurs a 10 percent uplift on model pricing for eligible models. The program is available to Enterprise customers, enabled per team, with a stated lead time of up to two weeks.

One configuration is generally available. US-only residency keeps inference, data processing, and storage of customer data in the United States. EU and Iceland coverage exists for inference only, on request. Broader EU support and APAC are described as in active development.

Enabling US-only residency restricts model selection to GPT models, Claude 4.6 and above, Gemini 2.5 Flash, Composer, and Grok 4.5. Grok 4.6 is not on that list. Gemini 3.1 Pro is not on that list. Selecting a model outside the list returns an error while residency is active.

The documentation also names what the residency guarantee does not extend to: single sign-on through WorkOS, codebase indexing when code is stored outside the region, bring-your-own-key, custom models reached through a base URL override or third-party gateway, MCP servers and external integrations including @Web, Bugbot and code review, links shared outside the team, and Cloud Agents triggered from Slack or the web. Buyers must map each of those data flows separately.

Sources: Cursor Docs, Privacy and Data Governance, read 15 Aug 2026 · Cursor Docs, Models & Pricing, read 15 Aug 2026

Enterprise commitment terms

The Pricing Policy was last updated on 21 May 2026, with a previous version dated 10 April 2026. Four provisions govern a signed term.

Precommitted Usage is the minimum usage spend named in the Order Form. Unused amounts do not roll into a subsequent term, and any unused balance remaining at the end of the Service Term is invoiced at that point. Only three fee types count toward it: Model API Fees, Cursor Token Fees, and Bugbot Usage Fees.

True Up is calculated on Active Users during the final 30 days of the True Up Period. Where that count exceeds the original seat baseline, Cursor invoices a Core License Fee plus a Precommitted Usage Fee for each additional seat, covering the remaining months of the term. Enterprise customers may reassign licensed seats to new users without additional charge, provided Active Users during the period do not exceed the seat count.

Model pricing is stated to be subject to change in accordance with the model provider’s list pricing. A commitment denominated in dollars therefore buys a quantity of tokens that the contract permits to move.

Amendments to the Pricing Policy take effect when posted. Cursor states that it notifies customers of material amendments by updating the date at the top of the terms and posting the latest version on that page, and that continued use confirms acceptance. No advance notice period was stated on the version reviewed.

Payment is net 30 from the invoice date. Purchase order documents must reach [email protected] at least five business days before the end of the next billing period, and Cursor states no obligation to submit invoices through a vendor management portal unless separately agreed in writing. The policy also grants rights to name the customer on Cursor’s website, in marketing materials, and at public events, with a review and consent step available only where the Order Form expressly provides one.

Sources: Cursor, Pricing Policy, 21 May 2026, sections 2.1, 2.4, 2.6, 3.2, 4.1, 4.2, 5.1, 5.2, 5.4, 8

The buyer decision path

Individual buyers. Do not read Pro Plus as a known threefold increase in Cursor Models capacity, or Ultra as a known twentyfold increase. Neither page maps the multiplier to that pool. Start monthly. Open the Spending tab at the beginning of a cycle, run representative tasks, and record what they consume before committing annually. Cursor’s own guidance places daily Agent use at $60 to $100 in total monthly usage, which sits above the Pro tier.

Team buyers. Configure the usage cap that the Pricing Policy grants administrators before inviting members. Decide whether the routed model stays hidden. If Cursor Router is enabled, budget for Balance and Intelligence at roughly twice Cost, and note that Teams and Enterprise are the only plans where the Cursor Token Rate applies.

Enterprise buyers. Ask for five items in the Order Form: the included Cursor Models quantity with its unit, the Other Models quantity, the overage rate and the control that caps it, the treatment of unused Precommitted Usage, and the notice mechanism for rate changes. The first of those is the one the public pages cannot supply.

Cursor also sells only through cursor.com. The pricing FAQ states that no resellers or third-party sellers are authorized and that subscriptions bought elsewhere may be suspended or terminated. Invoice and purchase order billing is documented on Enterprise. Buyers who cannot use a corporate card and cannot use a reseller must request Enterprise invoicing terms.

Sources: Cursor, Pricing, read 15 Aug 2026 · Cursor, Pricing Policy, 21 May 2026 · Cursor Docs, Models & Pricing, read 15 Aug 2026

Limitations of this audit

Five surfaces were outside scope and are named so that readers can weight the findings correctly.

No account was purchased, so the Spending dashboard was not opened and the starting denominator of either pool was not observed. No checkout screen or invoice was inspected, so the treatment of a routed model on a bill was not verified. No Order Form was reviewed, so Enterprise quantities and negotiated exceptions are unknown. Model behavior, routing decisions, and spillover order between pools were not tested. Locale review covered the English and Japanese versions of the pricing and model documentation only.

Statements that something was not located name the pages searched and the date. Cursor may publish figures in an authenticated dashboard, in an Order Form, or in a support channel that this briefing could not reach.

One ownership fact is recorded for currency rather than analysis. Anysphere, Inc. became a wholly owned subsidiary of SpaceX on 14 August 2026 through an all-stock transaction at an implied equity value of $60.0 billion.

Source: Bloomberg, 14 August 2026, reporting a Form 8-K filed by Space Exploration Technologies Corp.

FAQ

Methodology

This briefing reports the content of Cursor’s public documents and labels its inferences. It does not report tested product behavior.

Pages read on 15 August 2026: cursor.com/pricing across all six plan tabs and both billing intervals, cursor.com/docs/models-and-pricing, cursor.com/docs/cursor-router, cursor.com/terms/pricing, cursor.com/docs/enterprise/privacy-and-data-governance, cursor.com/changelog, and the Cursor help center article on usage and limits. Japanese versions of the pricing page, the Composer page, and the usage and limits article were also read. One external source is used, for the ownership line only.

Not reviewed: authenticated dashboards, checkout, invoices, Order Forms, support correspondence, and locales other than English and Japanese.

Prices and terms on this product changed repeatedly during 2026. Confirm current figures at checkout before purchase.

Verdict

Cursor publishes marginal rates with unusual precision and does not publish the individual-plan Cursor Models allowance those rates would convert into a monthly quantity.

That is workable for a buyer who can measure a cycle. It is not workable for a budget that must name a number before access. The public relative figures, three times and twenty times and five times, multiply a base that the reviewed pages do not state.

The action that resolves it costs $20 and one billing cycle. Buy monthly, capture the Spending tab at the start of the cycle, run the tasks the team actually runs, and read the denominator off the product rather than the marketing page. For an Enterprise term, require the quantity and its unit in the Order Form before signing.