Cursor Pricing (2026): $20 Buys Two Usage Pools

Cursor starts at $0, but the plan most working developers will consider is Pro at $20 per month. That price includes two usage pools, and neither pool has a fixed public dollar total on Cursor's current pricing pages. Every tier, rate, and limit below was re-verified against Cursor's live pricing and model docs on 24 September 2026.

Cursor lists its current plans on the official pricing page. The headline price is only the first layer: model choice, included usage, and on-demand billing decide the final bill.

Cursor pricing at a glance

Cursor Pro is the default paid choice at $20 per month. Stay on Hobby if occasional Agent and Tab limits do not interrupt you. Move to Pro+ or Ultra only when your usage dashboard shows a repeated need for their larger Agent limits.

The annual figures are monthly equivalents billed for a full year. That means $192 upfront for Pro, $576 for Pro Plus, $1,920 for Ultra, $384 per Teams Standard seat, or $1,152 per Teams Premium seat. Each annual option is 20% below twelve monthly payments.

Enterprise is not a volume-discount button. It is the route for pooled usage, invoicing, SCIM, access controls, audit logs, service accounts, and account support. Teams is likewise a governance purchase, not automatically a cheaper Pro plan.

Since September 23, 2026, Teams and Enterprise have also included access to Rollouts and Security Review. Rollouts monitors each deployed pull request and reports it as healthy, a regression, or inconclusive; Security Review looks for exploitable bugs. For the first 10 days after launch, Cursor includes trial usage credits for roughly 50 Teams changes or 500 Enterprise changes, but those temporary credits are not recurring plan allowances.

What Cursor Pro buys for $20

Pro buys editor features plus two monthly model pools. Treating the entire subscription as a bucket of requests misses both the included editor and agent tooling and the way model costs are metered.

The first pool is Cursor Models. It covers Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5. Cursor calls the included allowance “significantly more” usage, but its public docs do not assign that pool a fixed dollar, token, or request total. The pool may add substantial value, but a buyer cannot reproduce its exact capacity before subscribing.

The second pool is Other Models. It covers named third-party models at their listed API rates. Pro, Pro+, and Ultra include it, with on-demand billing available after the included amount is consumed. Cursor no longer publishes fixed dollar totals for these pools on its public pricing pages, so model selection and the usage dashboard matter more than a copied credit figure.

Both pools reset with the monthly billing cycle. Unused capacity does not become a bank for a later month. An annual subscription lowers the seat price, but it does not turn twelve monthly allowances into one transferable annual balance.

Paid individual plans also include unlimited Tab completions, extended Agent limits, frontier-model access, MCPs, skills, hooks, Cloud Agents, and Bugbot on usage-based billing. MCP, or Model Context Protocol, is the connector layer that lets an AI tool reach approved external systems and data. The feature list is broad, but model work still consumes a pool.

Cursor's own usage guidance is a better starting point than an old request count. It says daily Tab users typically stay within included usage, limited Agent users often stay within included usage, daily Agent users typically consume $60 to $100 in total monthly usage, and power users running multiple agents or automations often consume $200 or more.

For a solo builder who accepts Tab suggestions all day and opens Agent for a few focused refactors, Pro is enough until the dashboard says otherwise. For a founder delegating multi-file work to agents throughout the day, Pro+ is the more credible starting budget. Ultra belongs to the small group whose accepted work, not experimentation, consistently consumes power-user levels of model capacity.

Why the old break-even points no longer work

There is no reproducible public dollar threshold between Pro, Pro+, and Ultra. Cursor's current plan cards say Pro+ has 3x Pro limits on Agent and Ultra has 20x Pro limits on Agent, but the company does not publish the underlying dollar or token totals for either pool.

That removes the inputs needed for the old arithmetic. A guide cannot honestly claim that Pro+ wins at one exact spend or Ultra at another when the included pool values are not public. Pro+ also costs 3x Pro's monthly price for 3x its stated Agent limits, so the plan card alone does not imply a discount over Pro plus on-demand usage.

Ultra is a different tradeoff: it costs 10x Pro while advertising 20x Pro limits on Agent. That relative ratio can favor a sustained power user, but it is still not a dollar break-even because Cursor Models and Other Models are separate pools with undisclosed included totals.

Annual billing lowers seat prices by 20%, but it does not make the hidden pool amounts knowable. The defensible upgrade signal is repeated exhaustion of the relevant pool plus on-demand spend, measured across a normal monthly cycle.

If Grok 4.7, Grok 4.6, Grok 4.5, or Composer 2.5 handles most of your work, a lower tier may last longer because those models use the separate Cursor Models pool. If named third-party models dominate, watch the Other Models balance and on-demand line independently.

What one agent task can cost

A representative Claude Sonnet 5 task costs about $0.48 inside Cursor before any Team surcharge. The worked task uses 100,000 uncached input tokens, 400,000 cache-read tokens, and 20,000 output tokens. At Cursor's listed rates, that is $0.20 for input, $0.08 for cache reads, and $0.20 for output.

Cursor's public pages do not expose enough information to translate that task into a trustworthy task count for Pro, Pro+, or Ultra. A repository with more context, more tool calls, or longer output will also cost more than this worked shape.

The identical token shape on GPT-5.6 Luna costs about $0.052: $0.02 for input, $0.008 for cache reads, and $0.024 for output. The worked Sonnet task is about 9.2x that price. A model picker is therefore a budget control, not just a quality preference.

Cursor's free plan: who never needs to pay

Hobby is the current $0 entry point, but it is an evaluation plan rather than a dependable daily allowance. It requires no credit card and includes Composer, limited Agent requests, and limited Tab completions. Cursor does not publish a fixed numerical allowance on the current plan card.

The missing number matters. You cannot budget the free tier around a promised request count, and old fixed-request or unlimited slow-request claims describe earlier pricing systems. When Hobby's monthly Tab allowance is exhausted, the practical choice is to wait for the reset or pay for Pro's unlimited Tab completions.

You may never need to pay if you code occasionally, can tolerate the monthly limit, and use Cursor mainly to evaluate Composer or get intermittent help on a learning project. The free plan is also enough for someone whose primary editor remains elsewhere and who opens Cursor only for a narrow task.

You should pay when Tab suggestions or Agent access are part of a working day. A limit that interrupts paid development work costs more than the $20 subscription quickly, even before assigning a value to the extra models and Cloud Agents.

Cursor Start is a separate India-only option at ₹649 per month, tax included. It adds generous non-fast usage of Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5, plus Cloud Agents, Cursor for iOS, plugins, MCP servers, hooks, and skills. It does not include the Other Models pool, on-demand usage, Bugbot, Auto, Automations, or the Cursor SDK, and annual billing is not available.

Hidden Cursor costs and billing rules

The biggest billing risk is not the base price. It is allowing usage beyond the included pool without treating that setting as a budget decision. Cursor bills on-demand model use in arrears at the same listed API rates.

On-demand usage can turn a plan into a variable bill

Individuals can add on-demand usage after the included amount is consumed. Teams has on-demand usage enabled by default. Team admins can set a team-wide monthly spending limit, but per-member spend limits require Enterprise.

That difference changes the risk profile. A solo Pro user can decide whether to stop or continue. A Teams admin inherits a variable line item unless the organization configures the limit before broad adoption.

The Team token rate sits above model API pricing

Teams and Enterprise add a $0.25 per million token Cursor Token Rate to input, output, and cached tokens on eligible third-party requests. The rate also applies when any Auto mode chooses a third-party model, and it applies to BYOK, or bring your own API key.

On the 520,000-token Sonnet example above, the surcharge is $0.13 on top of the $0.48 model charge, for $0.61 total. Cursor's own Grok and Composer models are exempt. Eligible regional data residency adds another 10% uplift to model pricing.

Included model usage is not an external API wallet

The included usage lives inside Cursor's two plan pools. It is not a reusable Anthropic, OpenAI, or Google credit balance. Cursor resets both pools on the monthly cycle, and Team usage stays assigned to each user unless the company buys Enterprise pooled usage.

This is the app-versus-API split to remember: the plan pays for Cursor access and model work inside Cursor. External automation, a separate Claude Code API workflow, or another application's model calls need their own budget.

Annual savings come with prepaid exposure

Annual billing saves 20%, but the commitment is material: $192 for Pro, $576 for Pro Plus, and $1,920 for Ultra. Paid plans renew automatically unless cancelled before renewal.

Cursor's refund screen is narrow. A most recent individual subscription payment may qualify only when the charge is within 14 days and the subscription has not been used during that billing period. Approved refunds return in 5 to 10 business days. Outside that window, or after usage, the payment is generally not refundable. Consumed on-demand and fast-request charges are not refundable.

All general plan prices exclude applicable tax. Start's ₹649 price is the stated exception because it is tax inclusive.

Team seats can keep billing after a member leaves

Cursor bills active paid seats. Adding a member mid-cycle creates a prorated charge. Removing a member who already used credits keeps that seat occupied until the cycle ends, and some adjustments appear as a credit on a later invoice.

Teams Premium provides 5x Standard usage for 3x the price, and organizations can mix Standard and Premium seats. The ratio is attractive for a small number of heavy users; applying Premium to everyone wastes the point of mixed seats.

Cursor versus GitHub Copilot, Claude Code, and Devin

GitHub Copilot is cheaper for an extension-first developer, Claude Code is cleaner for terminal-first work, and Devin is the closest $20 alternative for an agent-centered desktop. Cursor earns its premium when the native editor, Tab, Composer, Cloud Agents, and model choice form one daily environment.

GitHub Copilot: the lowest entry price

GitHub Copilot costs $10 per month for Pro and currently includes $15 in total monthly AI Credits, unlimited code completion and next-edit suggestions, cloud-agent access, code review, and model selection.

Copilot Free includes 2,000 completions per month. Pro Plus costs $39 and includes $70 in total monthly AI Credits, while Max costs $100 and includes $200. One AI Credit equals $0.01, and paid usage can continue against a dollar budget.

The sharp comparison is Pro+: Copilot charges $39 and publishes a $70 total monthly credit allowance, while Cursor charges $60 and publishes the relative Agent limit but not the underlying dollar amount. Copilot wins if a disclosed credit budget and editor flexibility are the priority. Cursor wins if the integrated coding environment and Cursor Models earn the extra $21. The deeper product decision is covered in Cursor vs GitHub Copilot.

Claude Code: one subscription shared with Claude

Claude Code is included with Claude Pro at $20 monthly or $200 billed upfront annually, displayed as $17 per month. The subscription also covers the broader Claude product rather than a dedicated IDE.

Anthropic describes Pro as providing more usage than Free but does not publish one fixed message or token count on the current pricing page. Claude Code and the wider Claude product share the subscription, so the comparison turns on workflow rather than a reproducible per-task allowance.

Claude Pro is the better $20 purchase when terminal work and the wider Claude app both matter. Cursor Pro is the better $20 purchase when Tab, native editor workflows, multi-model choice, and Composer matter more. See Claude Code vs Cursor for the workflow decision beyond price.

Devin Desktop: the current Windsurf pricing destination

Devin Desktop, the new name for Windsurf, presents the plans reached from the live Windsurf page. Free costs $0, Pro costs $20, and Max costs $200.

The public plan overview does not state a numerical individual usage quota. Teams adds an $80 monthly base plus $40 per full developer seat. Devin matches Cursor's $20 and $200 individual anchors, but it does not beat Cursor on pricing transparency.

Choose Devin when its Desktop and Cloud workflow is the product you want, not because an unspecified quota looks cheaper. The acquisition-era product comparison is in Windsurf vs Cursor after Devin Desktop.

Which Cursor plan should you choose?

Choose the lowest tier that supports the work, then let accepted usage trigger an upgrade. A larger allowance is not valuable when inefficient context, unused capacity, or a poor model choice caused the bill.

Choose Hobby for evaluation, occasional coding, or a secondary editor. The plan works when waiting for a reset has no business cost.

Choose Start in India when Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5 cover the job and ₹649 is the right budget. Skip it when named third-party models, Auto, on-demand usage, Bugbot, Automations, or the SDK are required.

Choose Pro for most individual developers. The $20 tier already includes unlimited Tab completions, the paid individual feature surface, an included Other Models pool, and a separate Cursor Models pool.

Choose Pro+ when daily Agent use repeatedly exhausts Pro and its 3x Pro Agent limits solve a recurring bottleneck. Check which pool is running out before upgrading.

Choose Ultra for sustained power-user work that needs its advertised 20x Pro Agent limits. A one-off sprint may still be cheaper on a lower tier plus on-demand usage, so compare a full billing cycle before prepaying.

Choose Teams Standard for central billing, a team marketplace, shared context, usage analytics, privacy enforcement, and SAML/OIDC SSO. Mix in Premium seats only for the few developers whose Agent use justifies 5x the Standard allowance. Choose Enterprise for pooled usage, invoicing, SCIM, advanced controls, and per-member limits.

1. 1

   Start monthly

   Use Hobby for evaluation, then Pro on monthly billing for working use. Keep on-demand disabled until you understand the pool behavior.

2. 2

   Read both pools

   At the end of a full billing cycle, separate Cursor Models use from Other Models use. Do not upgrade because one combined total looks large.

3. 3

   Upgrade the exhausted pool

   Move to Pro+ when daily Agent work repeatedly exhausts Pro, and consider Ultra only for a sustained power-user pattern. Choose annual billing only after the workflow and model mix are stable enough to justify prepayment.

Cursor pricing FAQ

How much does Cursor Pro cost?

Cursor Pro costs $20 month to month or $16 per month equivalent on annual billing, which is $192 upfront. It includes separate Cursor Models and Other Models pools, but Cursor does not publish either pool as a fixed dollar total on its current public pages.

Is the Cursor $20 plan unlimited?

No, not for Agent and model use. Paid individual plans include unlimited Tab completions, but Agent work consumes the Cursor Models and Other Models pools. After included usage is consumed, you can enable on-demand billing or upgrade.

Is Cursor free?

Yes. Hobby costs $0, needs no credit card, and includes limited Agent requests, limited Tab completions, and access to Composer. Cursor does not publish a fixed numerical Hobby allowance on the current plan card, so free is best for light or interruptible use.

Does Cursor offer a student discount?

Cursor's live Students page did not list a standing student subscription discount when verified on 24 September 2026. It says anyone can start free and points students to promotions at on-campus and online events starting in fall. Older free-Pro offers should not be treated as current eligibility.

What is Cursor's refund policy?

The most recent Pro, Pro+, or Ultra payment may qualify when it was charged within 14 days and the subscription has not been used in that billing period. Outside that window, or after usage, it is generally non-refundable. Consumed on-demand and fast-request charges are not refundable.

Did Cursor pricing change in 2026?

Yes. Since 23 September 2026, Teams and Enterprise include access to Rollouts and Security Review, with temporary launch credits for roughly 50 and 500 changes during the first 10 days. Current individual pricing describes Pro+ and Ultra with 3x and 20x Pro Agent limits, while the public pages do not publish fixed dollar totals for the two included pools.

What is cheaper than Cursor?

GitHub Copilot Pro is cheaper at $10 per month and includes $15 in current monthly AI Credits. Claude Pro matches Cursor Pro at $20 monthly and includes Claude Code plus the wider Claude app. Devin Pro also costs $20, but its public plan overview does not quantify the individual usage quota.

Before adding another developer subscription, get the AI Business Workflow Audit Checklist.