Hobby is free, Pro is $20/mo, Teams Standard is $40/user, and Enterprise is custom. Cursor does not publicly publish prices for Pro Plus, Ultra, or Teams Premium; it describes them as 3x Pro, 20x Pro, and 5x Standard agent limits respectively. Paid plans carry two separate pools: generous included usage of Cursor’s own models (Grok 4.5 and Composer 2.5), and a dollar-denominated allowance for third-party models. Tab completion is unlimited and draws from neither. When the third-party pool runs out you continue on demand at the same API rates.
- Published prices: Hobby $0, Pro $20, and Teams Standard $40 per user. Pro Plus, Ultra, and Teams Premium prices are not publicly published.
- There are two pools, not one credit balance: Cursor Models (first-party) and Other Models (everyone else).
- The Other Models pool is denominated in dollars. Compare its allowance with the current price shown in Cursor’s plan picker.
- Tab completion is unlimited on every paid plan and draws from neither pool.
- Running out does not degrade you. You continue at the same API rates on demand, or upgrade.
- The single biggest lever on your bill is which model you pin, not how much you use Cursor.
The plans
Checked on cursor.com 7 August 2026. Annual billing is offered on the pricing page.
| Plan | Price | Included: Cursor models | Included: other models | For |
|---|---|---|---|---|
| Hobby | $0 | Limited | None | Evaluating it, no card required |
| Start (India only) | ₹649/mo, tax inclusive | Generous | $0 | Daily building on first-party models |
| Pro | $20/mo | Generous | $20 | Individual daily use |
| Pro Plus | Not publicly published | Generous | $70 | Heavy third-party model use; 3x Pro agent limits |
| Ultra | Not publicly published | Generous | $400 | All-day agent work; 20x Pro agent limits |
| Teams Standard | $40/user/mo | Generous | Per seat | Central billing, admin, policy |
| Teams Premium | Not publicly published | Generous | Per seat, 5x agent limits | Teams that live in the agent |
| Enterprise | Custom | Pooled | Pooled | SSO, SCIM, audit logs, procurement |
Cursor is built by Anysphere. Enterprise terms, pooled usage, and data-handling commitments are negotiated with them directly; the individual tiers were never the right comparison if those are requirements.
Two pools, not one credit balance
This is where nearly all the confusion lives. As of August 2026 Cursor splits usage into two separate pools, and they behave differently.
| Pool | What is in it | How it is metered |
|---|---|---|
| Cursor Models | Grok 4.5 and Composer 2.5, Cursor’s own models | Generous included usage on every paid plan |
| Other Models | Forty-plus third-party models from Anthropic, OpenAI, Google, Moonshot, Z.ai and others | Charged at each model’s API price against a dollar allowance |
Grok 4.5 is jointly trained by Cursor and xAI for long-running coding work; Composer 2.5 is Cursor’s own agentic coding model. The practical consequence is that a whole working style exists where you never touch the metered pool at all: Tab for editing, Composer for agent runs, and a third-party model only when the task genuinely needs one.
What draws from which pool
| Activity | Draws from |
|---|---|
| Tab completion | Nothing. Unlimited on Pro, Pro Plus, and Ultra. |
| Grok 4.5 or Composer 2.5 | The Cursor Models pool |
| Auto Cost | The Other Models pool, at a flat per-million-token rate regardless of which model runs |
| Auto Balance or Auto Intelligence | The Other Models pool, at the API rate of whichever model it routes to |
| Manually pinning a third-party model | The Other Models pool, at that model’s API rate |
| Max mode (legacy request-based plans) | Billed at the model’s API rate plus 20% |
Auto is not one thing any more
Older guides say "Auto is free". That was true of an earlier Cursor and is not true now. Auto has three modes and they price differently:
- Auto Cost is priced per million tokens at a flat rate no matter which model handles the request, and it is exempt from the Cursor Token Rate surcharge on Teams and Enterprise. This is the cheap, predictable setting.
- Auto Balance and Auto Intelligence route to better models when the task warrants it, and you pay that model’s API rate for the tokens it actually used.
The Cursor Token Rate
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 own price. It applies when you select a third-party model directly and when Auto Balance or Auto Intelligence routes to one. Auto Cost and all first-party Cursor models are exempt. On individual plans it does not apply.
What happens when the allowance runs out
Two options, and neither of them is a downgrade.
- Add on-demand usage. Keep going, billed monthly at the same API rates you were already being charged against the included allowance.
- Upgrade the plan. Move to a tier with more included usage, which is the cheaper answer if you exhaust the pool consistently rather than once.
The corollary is that a Cursor bill can exceed the sticker price without anything going wrong. If you want a hard ceiling, the lever is the routing default, not the plan.
What a month actually looks like
The third-party allowance is denominated in dollars, so the useful question is what draws it down. Three shapes of usage, and where each lands:
The Tab user
Autocomplete all day, Composer or Auto Cost for the occasional multi-file change, a frontier model only for something genuinely hard. The dollar pool barely moves. Pro at $20 is correct and will not run out.
The agent user
Several agent runs a day on a pinned frontier model, across a large codebase. This is the profile that exhausts $20 in the first half of the month. Compare Pro Plus in the current plan picker with Pro overage, or consider Ultra if it happens twice.
The long-context user
Big repository, long sessions, expensive models, repeatedly. Compare the current Ultra price with on-demand billing on top of Pro Plus. Cursor does not publicly publish the price, so use your signed-in plan picker and actual overage rather than a fixed crossover.
Teams, students, and the edges
Teams Standard at $40 per user
Teams Standard is Pro plus centralised billing, admin analytics, team-wide rules, and SSO. Teams Premium adds 5x Standard agent limits, but Cursor does not publicly publish its price. The relevant question for a small team is whether $40/user beats everyone expensing Pro at $20: it does not on price, it does on control. Buy it for the admin surface and the single invoice, not to save money.
Enterprise
Custom pricing with pooled usage, invoice and PO billing, SCIM seat management, access controls, audit logs, and priority support. If procurement and compliance are the binding constraints, this is the only tier that answers them.
Students
Cursor ran a free one-year Pro offer for verified students. As of August 2026 the students page does not offer it: it points students at the free Hobby tier and at promotions run through on-campus and online events. If you are planning a budget around a free student year, verify current eligibility on cursor.com/students before you rely on it.
Max mode
Max mode extends a model’s context window beyond its default and is billed at the model’s API rate plus 20%. It exists on legacy request-based plans; the current token-based Pro, Pro Plus, and Ultra plans do not present it. If a guide tells you Max mode is your problem, check which plan generation it was written for.
Cancelling
Self-serve in account settings. Access continues to the end of the paid period. If you are cancelling because the allowance ran out rather than because the product is wrong, try a month routed to Auto Cost and the first-party models first. That is a different problem with a free fix.
Why the pricing confuses everyone
Most of the contradictory advice online is not wrong, it is stale. Cursor has changed billing model more than once, and each change left a layer of write-ups describing a system that no longer exists.
The eras you will find described online.
| Era | How it worked | Why advice from it misleads |
|---|---|---|
| Fast and slow requests | A monthly count of "fast" premium requests, then a slow queue | People still ask "how many requests do I get". There is no request count now. |
| One credit pool | A single dollar-denominated pool that premium calls drew from | Closer, but it does not describe the first-party pool that sits beside it. |
| "Auto is free" | Auto and Tab stopped consuming the pool | Tab is still unlimited. Auto now has three modes and two of them are billed at model rates. |
| Two pools (current) | Cursor Models and Other Models, metered separately | This is today. The lever is routing, not volume. |
Cursor versus the subscription agents
Entry and top individual tiers, checked 7 August 2026.
| Entry | Top individual | What happens when you run out | |
|---|---|---|---|
| Cursor | $20 Pro | Ultra price not publicly published | On-demand billing at API rates, no quality downgrade |
| Claude | $20 Pro | $200 Max 20x | New turns stop until the rolling window refills |
| ChatGPT / Codex | $8 Go, $20 Plus | $200 Pro | Credits, priced per million tokens by model |
| GitHub Copilot | $10 Pro | $100 Max | Completions continue; agent work needs credits |
Four vendors, four different meters, all roughly $20 to get in and $100 to $200 at the top. The interesting comparison is not sticker price, it is what happens at the ceiling: Cursor and Codex keep working and bill you, Copilot keeps completing, and Claude makes you wait.
Making a plan last
- Default to the first-party pool. Composer 2.5 and Grok 4.5 have their own generous allowance and do not touch the dollar figure on your plan.
- Use Auto Cost when you want Auto. Flat per-token pricing regardless of which model runs, and no Cursor Token Rate on Teams.
- Reach for a frontier model deliberately. Hard debugging and architecture, not renaming variables.
- Use Tab more. It is unlimited and it is the feature Cursor is genuinely best at.
- Check your pinned model monthly. One forgotten setting is the most common cause of a surprise invoice.
- Compare on-demand against the next tier. On-demand is charged at API rates with no discount, so consistent overage almost always means the upgrade is cheaper.
Questions people ask
Hobby is free, Pro is $20/mo, Teams Standard is $40/user/mo, and Enterprise is custom. Cursor does not publicly publish prices for Pro Plus, Ultra, or Teams Premium; it describes them as 3x Pro, 20x Pro, and 5x Standard agent limits respectively. There is also a Start plan in India at 649 rupees per month. Checked August 2026.
Paid plans include a dollar-denominated allowance for third-party models: $20 on Pro, $70 on Pro Plus, $400 on Ultra. Cursor’s own models (Grok 4.5 and Composer 2.5) sit in a separate pool with their own generous inclusion, and Tab completion is unlimited on all paid plans.
There is a free Hobby tier with limited agent requests and no credit card required, which is enough to evaluate it. Sustained use needs a paid plan.
You either continue with on-demand billing at the same API rates, or upgrade to a plan with more included usage. Cursor documents that requests are not downgraded in quality or speed when you move to on-demand.
It depends which Auto you are on. Auto Cost is billed at a flat per-million-token rate regardless of the model chosen. Auto Balance and Auto Intelligence are billed at the API rate of whichever model they route to. Tab completion remains unlimited and free on paid plans.
Ultra has 20x Pro agent limits and Pro Plus has 3x, but Cursor does not publicly publish either dollar price. Compare the current figures in the plan picker with your actual on-demand overage before upgrading.
The automatic free Pro year is not offered on the students page as of August 2026. Students get the free Hobby tier plus promotions run at on-campus and online events. Verify current eligibility on cursor.com/students before planning around it.
Both start at $20. Cursor gives you an editor plus agent with a dollar meter that keeps working when it runs out. Claude gives you an agent with time-window allowances that stop and refill. For heavy autonomous runs Claude tends to go further per fixed dollar; for mixed editing Cursor stretches further because Tab and the first-party models sit outside the metered pool.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Cursor pricing plan tiers and prices
- Cursor docs: pricing included usage per pool, Auto modes, Cursor Token Rate
- Cursor docs: models first-party models, Max mode
- Cursor for students current student offer