Cursor pricing in 2026: every plan, both usage pools, and what it actually costs

Cursor is the most expensive developer tool company ever sold: SpaceX closed its $60 billion all-stock acquisition of Cursor’s parent Anysphere on August 14, 2026, a deal announced just days after the largest IPO on record and widely reported as the largest acquisition of a venture-backed startup on record.

And yet a Pro subscription still starts at $20 a month. That gap between a $60 billion price tag and a $20 sticker is the whole story of Cursor pricing: the usage system underneath the sticker is what your bill is made of.

We watch that system professionally. Our CTO published why he gives every CloudZero engineer a $5,000 monthly Claude Code token budget, with the math in the open, and we’ve shipped tooling that meters coding-AI spend alongside a Claude Code plugin for exactly this category. Per-engineer AI budgets are the fastest-moving line in engineering spend, and Cursor is where many of them live.

What are Cursor’s plans and prices in 2026?

Here are the current Cursor plans:

Annual billing saves roughly 20% across paid tiers. Features barely differ between Pro, Pro+, and Ultra; you’re buying pool size, which makes Cursor unusual among dev tools: the plan question is really a consumption forecast. Cursor Pro is the right starting point for most working developers, and Cursor’s own guidance is to upgrade only when included usage repeatedly blocks accepted work.

How does the Cursor credit system work now?

Cursor has repriced more than once. The eras, in order:

The current structure runs two pools: one for Cursor’s own models (Composer 2.5 and the Grok family, priced generously) and one for third-party frontier models, which draw usage at those providers’ API rates. On Pro the included pool roughly equals the $20 price.

Cursor describes the upper tiers by multiplier rather than fixed dollar amounts, and recent reporting places their included usage above the sticker, so read your plan’s pool in the dashboard rather than trusting any table, including this one, as the final word.

Once a pool is exhausted, on-demand usage bills in arrears, which is where surprise Cursor billing stories come from. The mechanics reward one habit above all: knowing your burn rate before the invoice does.

Which Cursor models cost the most?

The single biggest driver of your bill is not your plan; it’s which model handles each request.

The model rates are pass-through: Cursor’s own Composer 2.5 has been reported around $0.50 per million input tokens and $2.50 per million output, while manually selecting a third-party frontier model (Claude, GPT, Gemini) draws your pool at that provider’s own rates, the same rates on OpenAI’s rate card or behind Claude Code’s pricing.

A premium model with a huge context window can cost an order of magnitude more per request than a routed default.

Cursor’s own published estimates make the spread concrete: daily Tab users and light Agent users usually stay within included usage, daily Agent users typically land around $60 to $100 a month all-in, and power users running several agents often pass $200.

Same editor, 10x range, and the variable is model selection and agent intensity rather than the subscription. The practical rule: let Auto route the routine work, reserve manual frontier picks for problems that earn them, and treat Max Mode and background agents as the premium line items they are.

What happens when you hit a Cursor usage limit?

Running out mid-month is the most common Cursor complaint, and it is almost always a settings problem: the limits are dollar-denominated pools, not request counts, so “how much do I have left” requires looking. The usage dashboard shows the pool, the burn, and what consumed it; the things that drain it fastest are manual frontier-model selection, Max Mode, large contexts, and background agents left running.

On Teams, per-user analytics show which seats are heavy, which is how you decide who belongs on a $120 Premium seat and which seats have gone quiet, since an unused $40 seat is the easiest line item in the stack to find and to cut.

The Cursor cost conversation changes shape at the team boundary: an individual watches a dashboard; an organization needs the numbers somewhere finance can see them, which is the last section of this guide.

Is Cursor free?

Free or paid is a genuine choice here, within limits. The Hobby tier is a real free plan rather than a trial: limited Agent requests, limited Tab completions, no credit card, plus one week of full Pro to start. The free tier is built for evaluation and light personal projects, not daily work.

Anything resembling daily professional work hits the ceilings quickly, which is the honest answer to whether Cursor is free for real use: it’s free to adopt, and $20 is the practical floor once it becomes your editor.

SpaceX owns Cursor now: what changes for pricing?

Yes: Cursor is now owned by Elon Musk’s SpaceX, and the deal is done rather than pending: SpaceX secured an option in April 2026 (acquire for $60 billion, or pay a reported $10 billion for a partnership), exercised it on June 16, and closed on August 14, 2026, making Anysphere a wholly owned subsidiary two months after SpaceX’s record IPO and following its earlier absorption of xAI.

Cursor’s revenue had reached roughly $4 billion annualized by June 2026, about three-quarters of it enterprise, against a company founded in 2022, which is the growth story the $60 billion price was buying.

For budget owners, two practical implications. First, the Grok model family’s native presence in Cursor now has an ownership logic behind it, so watch routing defaults: which models Auto favors is now a vertically integrated decision, and routing is where your money goes.

Second, Cursor has not announced any pricing change since the close. Budget on the structure rather than today’s numbers, because this market has repriced four times in eighteen months.

How does Cursor compare with Claude Code and the field?

At the top end the comparison is direct: Cursor Ultra at $200 a month sits against Claude Code Max at the same price, an IDE against a terminal-native agent, and the right answer is workflow-dependent enough that we keep a dedicated Codex vs Claude Code comparison current.

Per-seat alternatives price differently again: GitHub Copilot’s enterprise seats are flat-rate, which trades Cursor’s usage ceiling risk for a utilization question instead. The pattern across the whole AI coding tools cost landscape: flat seats shift the risk to adoption, usage pools shift it to behavior. Neither is cheaper in general. They fail in different directions.

How do you track Cursor costs across a team?

Here’s what the dashboard can’t do: put Cursor next to everything else. A modern engineering org’s coding-AI spend spans Cursor seats and overages, Claude Code tokens, direct API keys, and gateway traffic, and each tool reports only itself. Engineering-AI spend is among the fastest-growing categories our 150+ AI statistics library tracks, and it’s the category most likely to live in five dashboards and zero budgets.

The fix is the discipline our CTO’s $5,000-per-engineer post models in public: give the spend a per-engineer denominator and a visible home. Cursor invoices and usage exports flow into CloudZero alongside cloud bills and token invoices through AnyCost custom sources, team and repo context arrives via the telemetry streams API, and ordinary AI cost management takes over: cost per engineer, per team, per agent workflow, trending against output.

That’s the difference between “the Teams invoice was $14,000” and “platform squad’s per-engineer AI cost rose 30% because two background agents never sleep.” One of those sentences gets a budget renewed.

The plan is a floor; the model picker is the price. Request a demo to see per-engineer coding-AI spend across Cursor, Claude Code, and your APIs in one place, get a free cloud cost assessment to find the AI spend hiding outside your cloud bill, or take the self-guided product tour at your own pace.

Cursor pricing FAQs

Author Spotlight

Lyne Carolyne

Lyne Carolyne has several years of experience in AI and cloud economics and brings that understanding into the content she creates. Outside work, she's an avid explorer.