Cursor AI Review 2026: Keep Pro at $20 or Move to Premium at $30?

Updated November 14, 2026

Premium costs $300 if you pay for a year, two months free; Cursor no longer lists yearly prices for its older plans; and OpenAI’s models stay in Cursor on a temporary extension with no new end date.

Since October 12, new customers can only buy Cursor through The X Subscription, so the Cursor AI review answer depends on one thing: whether you already pay. If you’re new, start on Premium at $30 a month, or $300 a year. If you already pay $20 for the older Pro plan and lean on Claude or GPT, keep it: switching is permanent and ends your included Claude and GPT usage.

⚡ TL;DR: The Bottom Line

What It Is: An AI code editor and agent workspace from Anysphere, now a SpaceX subsidiary. You chat with agents that edit your code, run them locally or in the cloud, and pick from Cursor’s own Composer 2.5, xAI’s Grok, Claude, GPT and Gemini models.

Best For: Developers who want an editor and agents in one place and are happy to watch a usage meter.

Price: Since October 12, new sign-ups get The X Subscription: Free, Plus $8 a month (no Claude or GPT), Premium $30, Super $100, Ultra $200, Teams from $30 per user. Existing subscribers keep Pro at $20, Pro+ at $60, Ultra at $200 and Teams from $40 per user.

Our Take: Already on Pro at $20? Keep it for now: your included Claude and GPT usage continues until you switch. Signing up today, Premium at $30 is the plan most people should start on, since xAI says it carries Cursor Pro’s benefits; Plus at $8 only fits if Grok models are enough.

⚠️ The Catch: The new plans give usage only as multiples (Premium is 4x Plus, Ultra 50x), not as a number, and on them every Claude or GPT request you pick is billed on top at API list prices.

What Cursor AI Is, and What Decides Your Bill

Cursor started as an AI code editor and has turned into a place to run coding agents, AIs that carry out a multi-step change for you. You open a folder, press Cmd I to open Agent and describe the change you want. Agents can run on your machine, in Cursor’s cloud, or on machines you host yourself.

What you pay depends less on the plan name than on the models you pick. On the new plans, think of a family phone plan. Your editor’s agents, Cursor’s cloud agents, Grok Bot and the Grok chat apps all share one allowance, and it refills every week.

Claude, GPT and Gemini sit outside that allowance. They work like roaming charges: you pay per use at the API list price, the price per token (a chunk of text a model reads or writes) that the model maker charges developers, from credit you load in advance.

The older plans count differently. They have two allowances that refill once a month, one for Cursor’s own models and one for everyone else’s. Either way, if most of your work runs on Claude, the plan price can be only part of your bill.

What Changed: A New Lineup, GPT on Borrowed Time and New Agents

Two changes since our June version touch your bill directly. The rest change how you work.

  • One subscription for Grok and Cursor. On October 12, 2026, xAI launched The X Subscription, one plan for Grok, Grok Bot, Cursor and X Premium. New customers have started on it since that day. The older Cursor plans are closed to new customers, but if you have one, you keep it until you switch or cancel.
  • GPT access is on a temporary extension. On August 28, OpenAI said it intended to wind down the contract that supplies its models to Cursor, with a proposed shutoff on November 12 and no future models. On November 12 the access was extended, temporarily and with no new end date. So if you use GPT-5.6 Luna, Sol or Terra in Cursor, they’re still listed, with no promise of how long; Cursor’s own pages don’t mention it.
  • New agent features. October and November brought several, all in Cursor’s changelog. The one that touches your plan choice is Build Live Apps, which builds an app in Cursor and shares it by link: it needs cloud agents, which on the new plans means Premium or higher.
  • A security notice. On November 13 Cursor’s blog posted “An update on our recent security incident”. If you use Cursor at work, read it in full there.

If GPT models are the reason you pay for Cursor, we’d hold off on a yearly plan until OpenAI gives an end date.

Cursor AI Pricing: Every Plan, Old and New

If you sign up today, you choose from The X Subscription: Free, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200. If you already pay for Pro, Pro+, Ultra or Teams, you can keep that plan and its price; the older lineup is further down.

The X Subscription: the plans new customers can buy

PlanMonthlyPaid yearlyUsage, compared with PlusClaude, GPT and Gemini
Free$0$0Limited (no number published)Not available
Plus$8$80 a yearBaseNot available
Premium$30$300 a year4x PlusAPI list price, from prepaid credit
Super$100$1,000 a year15x PlusAPI list price, from prepaid credit
Ultra$200$2,000 a year50x PlusAPI list price, from prepaid credit

Paying for a year costs ten months’ price, so you get two months free. Plus costs $80 instead of $96 and Premium $300 instead of $360. Super is $1,000 instead of $1,200, and Ultra $2,000 instead of $2,400: about 16.7% off. We’d still pay monthly first, and commit to a year only after you’ve watched your weekly usage for a few weeks.

You won’t find the yearly prices on Cursor’s main pricing page, which has no yearly toggle; they’re in Cursor’s plan docs and on xAI’s pricing docs. On these plans, prices exclude taxes such as VAT, and you need to cancel at least 24 hours before renewal to avoid another charge.

The usage column is a ratio, not an amount. Premium gets four times what Plus gets, but neither Cursor nor xAI says how much Plus gets: no dollar figure, no request count, no number of tokens. It’s like a coffee shop telling you the large is four times the small without ever showing you the small cup. How that weekly allowance runs down is in the usage section below.

Plus is the cheap way in, with real gaps. It has no Auto (Cursor’s automatic model picker), no cloud agents, no Bugbot code review and no Grok Bot. It can’t use third-party models, and it runs at fixed settings with Fast mode off.

The two Cursor pages also disagree on its models. The docs list Grok 4.7, 4.6 and 4.5 plus Composer 2.5, while the pricing page’s Plus card says “Only Grok models are available on this plan”. Until they agree, buy Plus only if Grok alone would do your job.

Teams are priced per seat, one paid place per team member. Each member gets a Premium, Super or Ultra seat at the individual price and weekly usage; you can mix tiers, but Plus isn’t sold for teams. Team plans add single sign-on (SSO), central billing, usage analytics and enforced privacy mode. Enterprise adds pooled usage, invoicing and SCIM, which adds and removes seats from your company directory.

Already on an older plan? What you keep

Nothing changes for you unless you choose to move. You keep your plan, price and features, and you can still move between the older Cursor plans without switching to the new lineup, unless you pay through the Apple App Store.

Older planMonthly priceIncluded usageClosest new plan
HobbyFreeCursor’s own models, limitedFree
Pro$20Both pools: Cursor’s models and third-party modelsPremium ($30)
Pro+$60More of each pool than Pro (no amount published)Super ($100)
Ultra$200More of each pool than Pro (no amount published)Ultra ($200)
Teams Standard$40 per userBoth poolsTeams, Premium seat ($30)
Teams Premium$120 per user5x the Standard allowanceTeams, Super seat ($100)
Cursor pricing page showing Hobby free, Individual at $20 a month, Teams at $40 per user and Enterprise custom
Cursor’s pricing page before the October 12 change, on monthly billing: Hobby free, the Individual card from $20 and Teams at $40 per user. Only existing subscribers can keep these plans now (screenshot, captured October 11, 2026)

Cursor no longer publishes a yearly price for these plans. Our earlier version listed $16 a month for Pro paid yearly, and that figure is on no current Cursor page. You can still move an older plan to yearly billing from your billing dashboard, but no page gives the price, so check there before you commit.

If you’re in India, you may be on Start, an older plan sold only there. It covers Cursor’s own Grok models at a fixed medium effort level, the setting for how hard the model thinks, with no third-party models, on-demand usage, Bugbot or Auto.

Switching to The X Subscription is optional and permanent: once you switch, Support can’t undo it. Cursor says each paid new plan includes at least twice as much coding usage with its own models as the old plan it replaces. But it drops the included Claude and GPT allowance, and twice an unpublished amount is still an unpublished amount.

At checkout, your old plan is cancelled with a prorated refund for unused time. The refund comes back separately, so you pay the new plan in full that day, and the money can take a few days to return.

How Cursor’s Usage Limits Work, and What Happens When You Run Out

The week you run out is the week your bill can jump, and the new and old plans handle it differently. On the new plans you get one weekly pool for Cursor’s own models: Grok 4.7, 4.6, 4.5 and Composer 2.5. Agent in the editor, the command line, cloud agents, Grok Bot and the Grok apps all draw from it.

The meter works like a fuel gauge with no numbers on it: you can watch it fall, but you can’t count the miles left. Models draw from the pool “at different rates”, and neither Cursor nor xAI publishes those rates, the pool’s size or a fixed reset day. Your own next reset date shows in the editor settings and on your Grok dashboard.

Pick Claude, GPT or Gemini yourself and the request never touches the pool. On Premium and up you pay for it from prepaid credit, money you load in advance like a prepaid card, at the API list price from the first request. The exception is Cost mode, the classic version of Auto, where Cursor picks the model for you. It’s the only automatic choice individual plans get, and it always draws from your weekly pool, whichever model it picks.

When the weekly pool runs out, you can wait for the reset, keep working on prepaid credit at the same per-model rates, or move up a tier. Auto Top-up refills your credit when it falls below a level you choose, and a monthly cap stops paid extra usage whatever credit you hold. Unused weekly usage doesn’t roll over, and top-up credit expires 365 days after you buy it. So set the cap before you switch on Auto Top-up.

If you’re on an older plan

On Pro, Pro+, Ultra and Teams, your usage sits in two buckets, Cursor Models and Other Models, and both refill on your monthly billing date. Pick a third-party model, or let Auto route to one, and the request draws from Other Models at that model’s list price. Watch subagents, the helper agents a main agent hands work to: one pinned to Claude bills Other Models even when your main chat runs on Composer.

When a bucket runs dry, on-demand usage keeps you going, billed after the month at the same per-model rates. On Pro, Pro+ and Ultra it stays off until you turn it on and set a monthly limit in your spending dashboard. On Teams it’s on by default, and Hobby and Start have none, so AI features pause until the reset.

Old plan vs new plan, row by row

QuestionOlder plansThe X Subscription
How often usage refillsMonthly, two poolsWeekly, one pool shared with Grok
Claude, GPT and GeminiIncluded on Pro, Pro+, Ultra and Teams, from the Other Models poolPrepaid credit at API list price from the first request (Premium and up)
After you run outOn-demand, billed after the monthPrepaid top-ups, or wait for the weekly reset
AutoA route to a third-party model draws Other ModelsCost mode draws the weekly pool, whichever model answers
Unused usageDoesn’t roll overDoesn’t roll over

One heavy month on Pro, then on Premium

Say you’re on the older Pro plan and use up both pools mid-month. Then you run 10 more Claude Opus 5.5 agent sessions, each reading 1 million tokens and writing 200,000. Opus costs $4 per million in and $20 per million out, so each session is $4 + $4 = $8. Ten sessions add $80 of on-demand to your $20: $100 for the month.

The same 10 sessions would cost $10 on standard Composer 2.5 and $60 on Composer 2.5 Fast. The $80 is twice the $40 step up to Pro+, but Cursor doesn’t say how much more Pro+ includes, so you can’t know whether upgrading would have covered it.

Now run the same month on Premium. The 10 Opus sessions come out of prepaid credit from the first request, so you pay $30 + $80 = $110, whether or not your weekly pool runs out. Any Claude work that Pro’s Other Models pool covered earlier in the month gets billed on top as well. If Claude does most of your heavy lifting, the older Pro plan is the better deal for as long as you keep it.

The vagueness has history. In June 2025 Cursor swapped Pro’s 500 requests a month for $20 of usage at API prices, plus unlimited Auto. On July 4, 2025 it published “Clarifying our pricing”, admitting the change wasn’t communicated clearly and offering refunds for unexpected charges from June 16 to July 4. Auto now draws from your allowance on both lineups, so treat any “unlimited Auto” you remember as history.

Composer 2.5 Pricing: Is Cursor’s Own Model the Cheap Option?

Per token, yes, if you pick the standard version. Composer 2.5 costs $0.50 per million input tokens and $2.50 per million output; Composer 2.5 Fast costs $3 and $15. Cursor’s launch post says fast is the default option, while the plan docs say Plus runs with Fast off. So check which one your model picker shows.

Composer 2.5 is built on Moonshot’s open-source Kimi K2.5, a model anyone can build on, and trained on 25 times more synthetic tasks, practice problems generated for training, than Composer 2. So with Composer you get Cursor’s own training on top of an open model, not Claude or GPT under another name.

Cursor blog post stating Composer 2.5 is priced at $0.50 per million input and $2.50 per million output tokens
Cursor’s Composer 2.5 launch post: the standard and fast prices, with fast as the default (screenshot, captured October 11, 2026)

Here’s how it compares with other models in Cursor, per million tokens, from Cursor’s model docs. On the older plans, these rates are what each request takes from your pools. On the new plans, Claude, GPT and Gemini cost these rates from the first request when you pick them, while Composer and Grok cost them only once your weekly pool is empty.

Model in CursorOlder plansNew plansInput (per 1M)Output (per 1M)
Composer 2.5Cursor Models poolWeekly pool$0.50$2.50
Composer 2.5 FastCursor Models poolWeekly pool$3$15
Grok 4.7Cursor Models poolWeekly pool$2$6
Claude Haiku 5.5 (up to 100k input)Other Models poolPrepaid credit$0.10$0.50
Claude Sonnet 5.5Other Models poolPrepaid credit$2$10
GPT-5.6 TerraOther Models poolPrepaid credit$2$12
Claude Opus 5.5Other Models poolPrepaid credit$4$20
Claude Fable 5.2Other Models poolPrepaid credit$8$40

Teams and Enterprise also pay $0.25 per million tokens on top of third-party prices. Models whose cached input, text they’ve seen recently, costs under $0.10 per million are exempt. That makes Haiku 5.5 exempt and Sonnet 5.5 not, and Cursor’s own models never carry the charge. If your team lives on Sonnet, budget for it.

List Price of a 1M-Input, 200k-Output Job in Cursor (USD)

The chart is our own arithmetic at Cursor’s list prices, with no caching: one job of 1 million input tokens plus 200,000 output tokens. Claude Fable 5.2, which replaced Fable 5.1 in Cursor’s list, costs double Opus for that job, so save it for the jobs you can’t hand to anything cheaper.

Haiku 5.5 is the cheapest bar at $0.20, but only while your requests stay at or under 100,000 input tokens. Above that it bills five times as much: $1.00 for the same job. On routine work, standard Composer costs an eighth of Opus for the same job, so judge its answers on your own code before you lean on it.

What $0.50 buys on the new plans

Cursor’s launch post: “Composer 2.5 is priced at $0.50/M input and $2.50/M output tokens”.

Cursor’s plan docs: on The X Subscription those rates “apply when you buy additional usage beyond the weekly pool”. Inside the pool, each model draws “at different rates”, and Cursor doesn’t publish them.

For you: on an older plan, $0.50 is what Composer takes from your allowance. On the new plans it’s the overage price, and Cursor doesn’t say how many Composer jobs a week Premium covers.

🆕 Cursor 3 and the Agents Window

Cursor 3 arrived on April 2, 2026 as “Meet the new Cursor”, and its centrepiece is the Agents Window. All your agents sit in one sidebar, local and cloud, including ones started from mobile, web, Slack, GitHub and Linear. A session can move from cloud to local when you want to edit and check it on your own machine, or from local to cloud so it keeps running when you close the laptop. A diffs view lets you review and edit changes, then stage, commit and manage pull requests, and there is an integrated browser and a plugin marketplace for MCPs, skills and subagents.

Since then Cursor has stacked more on top. Projects, in beta since September 10, gives a coordinator agent a long-running job such as a migration; the coordinator plans, delegates to other agents and brings the work back for you to check. Self-hosted machines keep tool execution inside your own network, and the iOS app can now watch and reply to agents running on your computer, which has to stay on and online.

⚔️ Cursor vs Claude Code vs Copilot vs Antigravity

The cheapest paid entry point in AI coding is still about $8 to $20 a month. What changes is whose models you get and how the limits are counted.

ToolFree optionEntry paid planNext step upModels
CursorFree (limited agent requests)Plus $8 (no Claude or GPT), Premium $30Super $100, Ultra $200Grok included; Claude, GPT and other third-party models at API list price (Premium and up)
Claude CodeNoClaude Pro $20 ($17 yearly)Max from $100Claude only
GitHub CopilotFreePro $10Pro+ $39, Max $100Varies by plan
Google Antigravity$0 individual planVia Google AI ProGoogle AI UltraA variety of agent models

For a head-to-head with Anthropic’s terminal agent, read Claude Code vs Cursor, which compares every plan. If you are weighing Google’s free agent IDE, our Antigravity vs Cursor comparison and Google Antigravity review cover the limits. GitHub’s credit system is explained in our Copilot Pro+ review.

⚠️ Limitations to Know Before You Pay

  • No published allowance. The new plans are sold as multiples of Plus (4x, 15x, 50x), the older ones as more of each pool than Pro. Neither has a dollar amount or request count behind it, so you learn your real limit from the usage meter.
  • Fast may be on. Cursor’s launch post calls fast the default for Composer 2.5, and Fast costs six times the standard version per token. Check your model picker if cost matters to you.
  • Extra charges on teams. Teams and Enterprise pay $0.25 per million tokens on top of third-party prices, and regional data residency adds 10%. Models with cached input under $0.10, such as Haiku 5.5, are exempt.
  • Local agents need your machine. Remote control from the iOS app only works while your computer stays on and online.

📬 Enjoying this breakdown?

Get honest AI tool analysis delivered weekly. No hype, no spam.

Subscribe Free →

👤 Who Should Pay for Cursor (and Which Plan)

  • Stay on the free plan if you want to try the editor and the occasional agent request before paying anything.
  • Signing up now? Start on Premium ($30). Over Plus it adds cloud agents, Bugbot and Grok Bot, and Cursor pitches it at new Agent users; Claude or GPT costs extra, at API list prices. Plus ($8) suits you only if Grok models alone cover your work and you don’t need cloud agents; Cursor’s docs also list Composer 2.5 for Plus, but its pricing page says Grok only.
  • Already on Pro ($20)? Keep it while you lean on Claude or GPT: your included third-party usage stops the day you switch, and the switch can’t be undone. The heavy-month example shows the gap in dollars.
  • Move up to Super ($100) or Ultra ($200) only if agents run for you most of the day. They carry 15x and 50x the usage of Plus, and Cursor aims them at daily Agent users and people running several agents. Existing Pro+ ($60) and Ultra subscribers can keep their plans.
  • Pick Teams if you need SSO, central billing or enforced privacy mode. If you start a team now, you pay the same per seat as individual plans ($30, $100 or $200), and each seat keeps its own weekly pool. Existing Cursor Teams plans keep their prices.
  • Look elsewhere if you need a hard, published usage number, only want Claude (Claude Pro includes Claude Code for the same $20) or want a free agent IDE (Antigravity).

Picture your team of five using 40 million input and 10 million output tokens of Sonnet 5.5 a month. Premium seats cost 5 x $30 = $150, the Sonnet use lists at $180, and the Teams token charge adds $12.50: $342.50 in all. The same five people on individual Premium plans would pay $330. So the $12.50 a month buys your team the admin features, not more usage.

🚀 Getting Started With Cursor

  • 1. Install. Download Cursor for macOS 12+, Windows 10+ or Linux. On Linux, Cursor’s quickstart prefers the apt or yum package (sudo apt install cursor, or sudo dnf install cursor from the yum repo) over the AppImage; the packages add desktop icons, automatic updates and CLI tools.
  • 2. Open a folder. Cursor’s quickstart suggests starting with a small task.
  • 3. Open Agent with Cmd I. Press Shift+Tab in the agent box for Plan Mode on anything bigger than a one-file change.
  • 4. Pick the model on purpose. Standard Composer 2.5 or Grok 4.7 for routine work; Claude or GPT models when you need them.
  • 5. Set a spend limit before you pay for extra usage: a monthly cap on the new plans, an on-demand limit on the older ones.

❓ Cursor AI Review FAQs

Q: Is Cursor AI free or paid?

A: Both. Free costs nothing and runs Cursor’s own models with limited usage; paid plans start at $8 a month for Plus, or $80 a year paid up front. Every price, old and new, is in the pricing section.

Q: What is Composer 2.5 pricing?

A: Standard: $0.50 in and $2.50 out per million tokens; Fast: $3 in and $15 out. Cached input, text the model has seen recently, is cheaper still at $0.20 per million, or $0.50 on Fast. On the new plans you pay these rates once your weekly pool runs out. The Composer section compares it with Claude and GPT.

Q: Is there a Composer 3?

A: No. Cursor’s model docs list Composer 2.5, released May 18, 2026, as the only Composer model.

Q: What is the Cursor AI release date?

A: Cursor first launched in March 2023, according to Contrary Research. The current major version, Cursor 3, was released on April 2, 2026.

Q: Cursor vs Opus: can I use Claude Opus in Cursor?

A: Yes. Claude Opus 5.5 is in Cursor’s Other Models pool at $4 per million input tokens and $20 per million output tokens, billed against your allowance on the pre-October 12 plans; on The X Subscription it is billed separately at API list prices (Premium and up). Standard Composer 2.5 costs one-eighth of that per token.

Q: Who owns Cursor now?

A: SpaceX. Cursor announced on August 14, 2026 that SpaceX had completed its acquisition of the company. Cursor still offers Claude, GPT and Gemini models alongside xAI’s Grok.

Q: Which Cursor plans include Grok Bot?

A: On The X Subscription, Premium, Super and Ultra include Grok Bot, as do Teams and Enterprise; Free and Plus don’t. It has no allowance of its own: Grok Bot draws from the same weekly pool as your Cursor agents. xAI says Premium carries the benefits of SuperGrok, Cursor Pro and X Premium+ (our SuperGrok price guide covers the Grok side). What Grok Bot does is in our Grok Bot review.

Q: What happens when I run out of usage?

A: You get a notification in the editor, and your requests aren’t downgraded in quality or speed. On the new plans you wait for the weekly reset, buy top-ups or upgrade. On the older plans you turn on on-demand usage; Hobby and Start pause until the reset. The usage section has the details.

Keep Pro or Start on Premium? Three Checks First

Cursor is still worth paying for if you want an editor and coding agents in one place. Which plan fits turns on three checks.

  1. Which models you really use. If most of your agent work runs on Claude, price it at API rates first. In our example, ten big Opus sessions made a $100 month on Pro and at least $110 on Premium.
  2. Monthly or yearly. A Premium year costs $300 instead of $360, but no Cursor page shows a yearly price for the older plans.
  3. Your renewal date. Cancel at least 24 hours before it to avoid another charge, and spend top-up credit within 365 days of buying it.
4/5
Editor’s Rating

Sources and how we checked: on November 14, 2026 we read Cursor’s pricing page, its plan docs and older-plan docs, model docs, billing and usage help, changelog and blog. The same day we read xAI’s X Subscription pricing and usage-limits docs, its pricing page, and OpenAI’s post on its Cursor decision. The Composer 2.5 launch post and The X Subscription announcement were read on November 11, 2026, and the Cursor 3, quickstart and June 2025 pricing posts were read on October 11, 2026.

Every price here is in US dollars, and the new plans’ prices exclude tax. The launch date comes from Contrary Research, a third-party company profile. Contrary Research’s profile was read on November 14, 2026, and the Claude, GitHub Copilot and Google Antigravity prices were read on October 4 to 10, 2026.

Tanveer Ahmad - AI Tool Analysis
Written by Tanveer Ahmad

Founder of AI Tool Analysis. Reads every pricing page, help article and footnote so you don’t have to. Covering AI tools since 2025. See how we research →

📬 Stay Updated on AI Coding Tools

AI coding plans change every few weeks. New models launch, prices move and limits change. Don’t miss the next change that affects your bill.

  • ✅ Honest Breakdowns: We read the pricing pages and docs first
  • ✅ Price Tracking: Know when tools get cheaper or go free
  • ✅ Comparison Updates: As the market shifts, we update our verdicts
  • ✅ Feature Launches: Major updates covered within days
  • ✅ No Hype: Just the AI news that actually matters for your work

Free, unsubscribe anytime.

Want AI insights? Sign up for the AI Tool Analysis weekly briefing.

Newsletter

Signup for AI Weekly Newsletter

AI Tool Analysis newsletter preview showing weekly AI tool reviews

Last Updated: November 14, 2026

Cursor Version Covered: Cursor 3 with Composer 2.5 and Grok 4.7 (changelog through November 6, 2026)

Next Review Update: December 14, 2026

Have a tool you want us to cover? Suggest it here | Questions? Contact us