Cursor AI pricing in 2026: every plan, the credit system, and what it actually costs

On July 4, 2025, while the rest of the industry was grilling things, the CEO of one of the fastest-growing software companies in history spent his Independence Day publishing an apology about a price change.

The change had sounded small. Cursor’s Pro plan quietly shifted from “500 requests a month” to “$20 of usage a month.” The fallout wasn’t small: developers who had never once thought about tokens hit their limits in hours, some woke up to surprise bills in the hundreds, and the anger was loud enough that a full year later, “why is Cursor so expensive now” is still one of the questions Google suggests under this search.

Here’s the part almost nobody explains, and it makes the whole saga make sense. Cursor wasn’t gouging anyone. Per The Information’s reporting, its gross margin was negative 23% in the quarter ended January 2026, at a company then approaching $2 billion in annualized revenue. It was losing money on every dollar developers spent, because the frontier models underneath cost more than the subscriptions above. Those margins have since turned positive, and the repricings are why.

Marty Kausas, CEO of the support platform Pylon, summed up the entire category this year: “Nobody has figured out how to price AI.”

And yet: in June 2026, SpaceX agreed to acquire Anysphere, Cursor’s maker, for $60 billion in stock, the largest startup acquisition on record, with Cursor’s revenue at roughly $4 billion annualized and about 75% of its enterprise. The company that couldn’t price its own product profitably in early 2025 became the most expensive startup purchase in history eighteen months later. The pricing turbulence you’ve read angry threads about? That was the fix working.

Cursor has spent 2026 figuring it out in public: a two-pool credit system in June, its Bugbot agent repriced from $40 a seat to roughly $1-1.50 a run, an India-only plan priced in rupees. Every change follows the same logic, prices converging on actual usage, and every change breaks every guide written before it.

This article is built to do two jobs: give you every plan and price, and explain the system underneath well enough that the next change costs you exactly zero dollars of surprise.

What are Cursor’s plans and prices in 2026?

The full menu first:

Note: Cursor has changed its pricing four times since June 2025, so check the date on every pricing guide you read, including this one. Prices here were verified against cursor.com/pricing as of Sept. 2, 2026.

A new-user note that saves real money: there’s a one-week Pro Cursor free trial for new accounts. Run it during a real sprint, not a quiet week, because the entire art of choosing a tier is learning your usage shape, and a trial spent on toy tasks teaches you nothing.

One buying note before the mechanics: whichever tier a developer picks, the organization inherits the aggregate, which is why the teams that come out ahead treat plan selection as a per-developer visibility question, not a procurement checkbox. More on that machinery below.

Prices are the easy half. The system underneath is where the confusion, and the money, lives.

How does Cursor’s credit and usage system work now?

Plainly, because this is the part the confused-forum genre exists for, and because it changed again in June 2026.

Cursor credits are a monthly usage allowance, measured in what the underlying models actually cost, that every paid plan includes. That allowance is split into two pools. Cursor Models carries generous included usage for Grok 4.6, Grok 4.5, and Composer 2.5, all now first-party after the SpaceX deal. Other Models is the pool for third-party frontier models from Anthropic, OpenAI, and Google, charged at each model’s API price. The dashboard shows how close you are to each limit separately, a direct fix for the old single-meter problem.

Two changes older guides miss. Auto no longer bills as a free first-party request: every Auto mode bills at the list price of whichever model it routes to. And on Teams and Enterprise, third-party requests carry a Cursor Token Rate of $0.25 per million tokens on top of model API pricing, which first-party Grok and Composer requests avoid.

The underlying mechanics haven’t changed. Different models burn credits at wildly different speeds: a quick edit on a small model costs pennies of allowance, and a long agent session on a third-party frontier model can eat dollars of it in one sitting. Max Mode, which older guides warn about, now exists only on legacy request-based plans, billed at the model’s API rate plus 20%. If you signed up recently, you do not have it. Cursor usage past a pool’s limit either stops (if you’ve set a cap) or bills on at model rates (if you haven’t).

Three consequences worth tattooing somewhere. First, two developers on identical Pro plans can have completely different months: one never sees the ceiling, the other hits it by the 9th. Second, Composer 2.5 delivering near-frontier performance from the generous pool means “default to first-party, escalate deliberately” is now the largest single lever on your Cursor bill.

Third, set the spending cap before you need it. The people in the surprise-bill stories weren’t careless. They had default settings.

The two-pool design is, quietly, Cursor admitting what cost platforms like CloudZero have argued for years: AI spend is only manageable when it’s visible per meter, per user, while it’s happening. Cursor built that view for Cursor. The rest of the AI stack still needs it.

Which brings us to the question this search practically vibrates with.

Why is Cursor so expensive now?

Because the thing underneath it is expensive, and 2025 was the year that stopped being hidden.

Cursor’s old flat subscription worked when models were cheaper and sessions were shorter. Then agents arrived, sessions grew far longer, frontier models stayed pricey, and Cursor found itself selling dollars for roughly eighty cents at massive scale.

The June 2025 usage shift, the apology, the Bugbot repricing, June 2026’s two pools: it’s all one event repeating. A company converging on usage-based pricing, because seat pricing breaks when an AI does wildly variable work per user.

And the epilogue vindicates the pivot: by mid-2026 revenue had quadrupled to roughly $4 billion annualized, three-quarters of it enterprise, and SpaceX priced the whole company at $60 billion. Usage-based pricing didn’t just stop the bleeding; it built the thing a rocket company paid a record price for. Expect every AI tool you use to study this playbook.

The honest read for buyers: Cursor isn’t uniquely expensive; it’s unusually early to a cost structure every AI tool shares. The Cursor price you pay tracks the work you generate.

So the winning move was never finding a cheaper tool. It’s knowing your usage, fitting the plan to it, and re-checking quarterly, because in a market that’s repriced this often, even the right answer has a shelf life. The developers who never end up venting in forums aren’t the lucky ones. They’re the measured ones.

Is SpaceX buying Cursor, and will prices change?

SpaceX has already bought it. The acquisition, announced June 16, 2026 and completed August 14, 2026, is the largest startup purchase on record: $60 billion, all stock, for a company whose last private round in November 2025 valued it at $29.3 billion. The headline reads strangely until you see the revenue curve underneath it: roughly $500 million to $4 billion annualized in about 12 months, three-quarters of it enterprise.

For pricing, the practical answer. Plans, pools, and rates on this page reflect post-close pricing, and nothing has been announced since. The enterprise tilt of the revenue suggests continuity rather than shock, because roughly $2.6 billion in annualized enterprise revenue is not something an acquirer breaks on purpose.

The deeper signal for anyone budgeting AI tools: the Cursor SpaceX deal prices developer-tool spend as strategic infrastructure now. Which means the line item is permanent, growing, and worth measuring like one, the same conclusion finance teams keep arriving at from the cost side of the same table.

How much does Cursor actually cost per month?

Translate allowances into real months, with ranges because variance is the entire point:

Read the middle rows twice, because the Pro-to-Pro+ gap is where most of the internet’s Cursor anger actually lives: developers whose real usage is a $70 month sitting on a $20 plan, experiencing the difference as betrayal rather than arithmetic.

The Cursor AI pricing plans logic, in one line: buy the allowance your last month’s usage says you need, not the one your ambitions suggest. Revisit monthly. Downgrades carry no penalty and your usage data is free.

For Cursor Pro pricing specifically: real daily developers outgrow $20 fast, and the viral verdict that Pro “runs out of credits in two hours” under heavy agent use is exaggerated for normal work and dead accurate for frontier-model marathons.

The Cursor Pro business pricing step to Teams buys admin visibility more than raw capacity, and past ten seats, that visibility is the feature. And Cursor Ultra at $200 is what every AI tool’s top tier is: a bulk discount for people whose metered usage would cost more, wearing a subscription costume.

One elephant, named honestly: the open-source crowd (opencode and friends, bring-your-own-API-key, 160K+ GitHub stars of pricing frustration) exists because of this ladder. It’s a real option for terminal-native developers.

The tradeoff is eternal: you save the markup, inherit the plumbing, and still pay the model providers, whose rates are the actual floor under this entire market.

Is Cursor free, and what do students get?

Is Cursor free? Partially, and the free layer is bigger than most people realize.

The Cursor free plan (Hobby) includes limited completions and agent requests: enough to judge the editor, not enough to work in daily. The 1-week Pro trial covers the real-workflow test. And the student angle, corrected: Cursor runs no university-partnership program and no standing student discount. Its students page says anyone can start free, and that upgrade access comes through promotions at Cursor’s on-campus and online events. If you are a student, free Hobby access plus event promotions is the actual path. There is no institutional benefit to go and ask your school about.

The Cursor free layer’s honest ceiling comes straight from the economics above: every free request costs Cursor real model spend, so free tiers stay thin by necessity. Free gets you a verdict. Work needs a plan.

The free-tier math also previews the organizational truth: what’s free per developer is never free per company, because evaluation seats become production seats without a memo, which is how AI tooling lines grow 40% before anyone approves anything. Visibility from seat one is cheaper than surprise at seat fifty.

And speaking of plans priced for reality: the India Start plan at ₹649/month, priced in rupees with UPI and tax included, is the most interesting move Cursor has made all year, a bet that the next ten million AI developers price in local currency. If it works, expect regional pricing to spread across the whole market.

How does Cursor pricing compare with Claude Code and the field?

They’re two halves of the same budget line now. Cursor ide pricing wraps model costs in editor-shaped allowances; Claude Code pricing wraps them in tiered subscriptions or raw API metering. The ladders rhyme (Pro/Pro+/Ultra against Pro/Max), and both converge on the same truth from opposite directions: light users fit small tiers, agent-heavy users fit big ones, automation belongs on metered plans with caps.

The difference is what each plan buys: one Cursor plan burns allowance across many providers’ models, with GPT-5.6 Luna at $0.2 in and $1.2 out per million and Gemini 3.7 Flash at $0.75 and $3.5 among the budget-friendly picks, and Composer 2.5 at $0.5 and $2.5 on the Cursor Models pool. Claude Code is Anthropic’s models at Anthropic’s rates. GitHub Copilot plays the flat-rate seat game one door down.

Most engineering organizations now run two or three of these at once, which is precisely why the AI coding line item needs its own cost discipline, and why the head-to-head lives on its own page: Codex vs. Claude Code, with the raw model rates underneath everything in the LLM API pricing comparison and the AWS-side equivalents in the Bedrock pricing guide.

How do you keep Cursor costs under control?

Five moves, in payoff order, and the first takes ninety seconds.

Individual optimization only goes so far; across a 25-person team you need centralized visibility into who’s burning which pool on which models, and a 50-seat plan decision made on sticker prices is a guess. One made on a month of real usage data is arithmetic.

How do you track Cursor costs across a whole team?

Cursor’s own per-user dashboard (Teams and Enterprise) is the right first stop, and since June it splits the two pools cleanly. Its limit is its scope: it sees Cursor.

The actual problem is bigger. AI coding spend now spreads across Cursor, Claude Code, Copilot, and raw API keys nobody reconciles, with enormous usage variance between similar engineers, sitting next to the cloud bill those engineers also drive. Finance sees one line climbing. Nobody sees why.

CloudZero tracks Cursor spend at the account and team level and pulls it into the AI Hub alongside Claude Code, Copilot, every major AI provider, and your cloud spend. One allocated view, per developer, per team, per project, through the AnyCost pipeline that ingests any billing source.

Engineering leaders get the cost context surfaced where they already work, inside the coding agents themselves, while budgets put forward numbers on the AI tooling line and anomaly detection flags the runaway-session pattern while it’s still small.

The payoff is the sentence finance has been trying to earn the data for. Not “the AI tools line grew 40% again,” but “these teams’ AI tooling returned measurably more than it cost, buy more of it.”

In CloudZero’s 2026 AI ROI survey of 260 finance leaders, 87% say they need to tie AI spend to business outcomes within the year, and only 22% can today. AI coding tools are where that gap lands first: the most seats, the most variance, and the least tracking.

Bring your AI tooling bills to a demo and watch Cursor, Claude Code, and cloud spend allocated per developer, live.

Frequently asked questions about Cursor AI pricing

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.