Finding out whether the editor suits you
Pro
You have used VS Code for years and want to know whether the agent and the autocomplete are worth changing editors for.
$20/mo
$16/mo billed annually
Start on the free Hobby tier, not on this row: it needs no card, and the refund rules make a paid trial a poor idea. The most recent payment is refundable within fourteen days only if you have not used the subscription at all in that period, so a month of Pro bought to try things out is a month you keep paying for. When Hobby's unnumbered limits stop you, take Pro monthly rather than annually — the annual discount is a fifth, and it buys nothing back if you leave.
A full working day with the agent
Pro+
The agent is open most of the day, on a frontier model, doing real edits rather than answering questions.
$60/mo
$48/mo billed annually
The vendor's own recommendation for daily agent users, and its own figure for what that user spends in total is printed further down this page. Read the two together: if a daily agent user typically spends more than this tier costs, the difference arrives as on-demand usage, which on an individual plan stays off until you switch it on. Leaving it off turns the allowance into a hard stop — which, with no published size, is the only way to learn where the allowance actually ends.
Several agents at once, or cloud agents overnight
Ultra
You run parallel agents, automations or cloud agents that keep working while you are away from the keyboard.
$200/mo
$160/mo billed annually
Twenty times Pro's agent limits, and the tier whose published allowance most exceeded its own price before the figure was taken down — the archived documentation put it at double. Cloud agents bill at the selected model's API price and ask for a spend limit on first use, so the limit you set there is the real ceiling. Check which variant is selected, too: Cursor's own Composer runs in its fast variant by default, at six times its standard rate.
A team of eight
Teams Standard
A small engineering team wants central billing, SSO and shared rules without a sales process.
$40/seat/mo
$32/mo billed annually
Self-serve per seat, with SAML and OIDC included rather than sold separately. Three settings to change on day one. On-demand spending is on by default here, so set a team limit before anyone runs an agent overnight. Every third-party request carries the per-token platform fee printed in the table, even on your own key. And Grok Bot is enabled for every member with no switch to turn it off. Mix in Premium seats only for the people who actually hit the ceiling: five times the usage at three times the price.
An organisation with procurement and a security review
Enterprise
You need invoicing, pooled usage, SCIM, audit logs and a contract your legal team has read.
Custom
The only route with pooled usage, and the only one whose price is not published. Read the Pricing Policy before the order form: precommitted usage does not roll over and is invoiced if unused at the end of the term, seats are trued up against everyone who logged in during the last thirty days of each period, and by default the policy lets the vendor name you as a customer in its marketing. Privacy Mode is on by default here, and US-only data residency adds a tenth to model rates.
You already pay a model provider
Your own API key
You hold an Anthropic, OpenAI or Google key with committed spend or credits, and want Cursor's editor on top of it.
At model rates
On an individual plan this works cleanly: your provider bills the tokens and nothing is drawn from the included allowance. Three limits apply. Keys cover chat models only, so Tab still runs on Cursor's models. Requests still pass through Cursor's backend, and its zero-retention agreements do not cover them. And the documentation describes the key as an addition to a paid plan; whether the free tier accepts one is not stated.