Cursor in 2026: Pricing, Free Plan, Models and Limits

Cursor is an AI code editor and coding agent: you describe a task, and it reads your project, edits the files, runs commands and hands you the result to review. There is a free Hobby plan; paid plans run from $20 to $200 a month for individuals and start at $40 per user for teams (Cursor pricing, checked Sept. 25, 2026). It’s built for developers who want to delegate whole tasks to AI, not just get autocomplete.

This review is current as of September 25, 2026. It covers what changed since 2025, features, models, plans and usage limits, code privacy and alternatives, based on Cursor’s official pricing page, docs, blog and changelog, linked throughout.

What Changed Since 2025

Three big shifts since the first version of this page (May 2025):

• Cursor is now owned by SpaceX. Cursor announced its acquisition by SpaceX on August 14, 2026 (Cursor blog); Cursor did not disclose the price. SpaceX had already absorbed xAI, renamed SpaceXAI in July 2026. The visible result: Grok models now take center stage in Cursor.
• A new agent-first interface. Cursor 3, released April 2, 2026, adds the Agents Window, built from scratch, alongside the classic VS Code–based editor, which is still available (Cursor 3 announcement).
• Usage-based billing. The old request counts (2,000 completions a month on the free plan, as cited in the 2025 version of this page) have given way to an included usage amount, drawn down at each model’s price (Models & Pricing, checked Sept. 25, 2026).

Cursor has also ended its referral program, so this page contains no affiliate links.

What Does Cursor Do?

Cursor now pitches itself as a “coding agent” more than an editor. Documented features:

• Agent: describe the task in plain English; the agent explores the codebase, plans, edits multiple files, runs terminal commands and shows the diffs for you to approve. A Plan mode drafts a plan first.
• Tab: Cursor’s predictive autocomplete, unlimited on paid plans.
• Agents Window: run several agents in parallel across repos, locally, in isolated Git worktrees, over SSH or in the cloud, and hand a session off between local and cloud (Agents Window docs).
• Cloud agents: agents that work on their own virtual machine, build, test and demo the result, even with your laptop closed. Since September 2, 2026, they can also run on machines your company manages (changelog).
• Projects (beta since September 10, 2026): for large pieces of work (a feature, a migration, a full app), a coordinator agent delegates to subagents, keeps context over time and can be triggered by a Slack channel, a schedule or a pull request (Projects announcement).
• Bugbot: automated pull request reviews on GitHub. Teams and Enterprise plans also get Rollouts (tracks a change all the way to production) and Security Review (flags exploitable bugs on every PR), both launched September 23, 2026 (announcement).
• Everywhere you work: command line (Cursor CLI), Slack, GitHub, Linear, iOS app and web. MCP servers, skills, hooks and plugins connect other tools.
• Grok Bot: a general-purpose agent from SpaceXAI, distributed and billed by Cursor, that works on its own cloud computer and uses your apps. It’s included, with weekly usage that grows by tier, in Pro, Pro+, Ultra and Teams (Grok Bot help, checked Sept. 25, 2026).

The editor is still a VS Code fork, so your VS Code extensions, themes and keybindings carry over. Cursor runs on macOS, Windows and Linux.

Which AI Models Does Cursor Use?

Cursor splits usage into two monthly pools (Models & Pricing, checked Sept. 25, 2026):

• Cursor Models: Grok 4.7, 4.6 and 4.5 (SpaceXAI) plus Composer 2.5, Cursor’s in-house model. This pool is significantly more generous.
• Other Models: models from Anthropic (Claude Opus 5.5, Sonnet 5, Fable 5.1), OpenAI (GPT-5.6 Sol, Terra, Luna), Google (Gemini 3.1 Pro, Gemini 3.8 Flash) and others, charged at their API price.

Grok 4.7, introduced September 21, 2026, is the model Cursor features across its site. SpaceXAI reports 46.3% on CursorBench 4.0 and 71.0% on DeepSWE v1.1 (SpaceXAI announcement); these are vendor-published scores. Composer 2.5 (May 18, 2026) is the cheapest of Cursor’s own models (Composer 2.5 announcement).

Auto picks the model for you, and each request is billed at the price of the model it lands on. Its new version, Cursor Router, with three modes (Cost, Balance, Intelligence), launches on Teams and Enterprise first and reaches individual plans “a few months after launch” (Cursor help, checked Sept. 25, 2026).

Prices per million tokens, drawn from your included usage (Models & Pricing, checked Sept. 25, 2026). “Fast” variants cost more.

How Much Does Cursor Cost?

Prices from Cursor’s pricing page (checked Sept. 25, 2026). They’re in US dollars and exclude applicable taxes.

Cursor recommends Pro+ for daily agent users and Ultra for power users. Subscriptions are sold only on cursor.com; Cursor doesn’t authorize any resellers.

The old student deal (a free year of Pro) closed to new sign-ups on June 25, 2026. Cursor now offers credits at campus events and a request form for graduate students, researchers and educators (Cursor help, checked Sept. 25, 2026).

Is Cursor Free?

Yes, on the Hobby plan, with no credit card required. It gives you Agent, Chat and Tab completions with the Auto model, plus Composer (Cursor help, checked Sept. 25, 2026). Cursor doesn’t publish the limits: the pricing page only says “Limited Agent requests.”

That’s enough to try Cursor on a side project, not to work with the agent every day.

How Cursor’s Usage Limits Work

Every paid plan includes a monthly usage amount, and Cursor doesn’t publish how much. What is documented (Usage and limits, checked Sept. 25, 2026):

• Usage is drawn down at the API price of the model you use: a task on Claude Fable 5.1 burns through it far faster than one on Composer 2.5.
• Cursor’s own models (Grok, Composer) have a separate, larger pool.
• Usage resets every billing cycle; unused usage doesn’t roll over.
• When you hit the limit, the editor notifies you: turn on on-demand usage (billed monthly in arrears at the same rates) or upgrade. Requests are never downgraded in quality or speed.
• On individual plans, requests made with your own API key don’t draw from either pool; your provider bills you directly.

Cursor’s own ballpark figures (Models & Pricing): daily Tab users and light Agent users usually stay within the included amount; daily Agent users typically land at $60–$100 a month in total usage, and power users running multiple agents or automations often reach $200+. On Teams and Enterprise, third-party models also carry a “Cursor Token Rate” of $0.25 per million tokens.

You can track usage in real time on the Spending tab of your dashboard.

Code Privacy: What Happens to Your Code?

It all comes down to Privacy Mode. According to Cursor’s data use policy (updated September 3, 2026):

• Privacy Mode on: your code isn’t used for training by Cursor or its model providers, which operate under zero data retention agreements. Data may be kept if a request trips abuse detectors.
• Privacy Mode off: Cursor may use and store code, prompts and editor actions to improve its features and train its models.

Privacy Mode is on by default for team members and Enterprise; individual users can turn it on in settings. One exception: Claude Fable 5.1 and Fable 5 require Anthropic to retain inputs and outputs (for safety reviews, not training), so with Privacy Mode on, these models need explicit approval before use (Privacy and Data Governance, checked Sept. 25, 2026).

On compliance, Cursor lists a SOC 2 Type II attestation plus ISO/IEC 27001, ISO/IEC 42001 and AIUC-1 certifications (Security page). US-only data residency is available to Enterprise customers at a 10% uplift on model pricing; EU + Iceland coverage is inference-only and on request.

Cursor’s Drawbacks

• Opaque limits. No plan states its included usage in dollars or requests; your real cost depends on the models you pick and can exceed the subscription once on-demand usage is on.
• Cloud-dependent: models run on Cursor’s or its providers’ servers, not locally.
• A future tied to SpaceX: Grok and Grok Bot are increasingly central to the plans. Cursor hasn’t made a public commitment on keeping rival models, which were all still available as of Sept. 25, 2026.
• You still need to review everything: like any coding agent, Cursor can write wrong or brittle code. Review proposed changes before merging.

Cursor vs. GitHub Copilot, Claude Code and Mistral Vibe

Sources, checked Sept. 25, 2026: GitHub Copilot, Claude, Mistral and Cursor pricing.

• GitHub Copilot fits if you don’t want to switch editors: it plugs into VS Code, JetBrains, Xcode or Neovim and costs half as much to start.
• Claude Code fits if you already pay for Claude: it’s included in Pro and shares that plan’s usage limits.
• Mistral Vibe, in Code mode, suits developers who prefer a European provider with one subscription for work and coding: see our Mistral Vibe guide.
• Cursor wins when you want several agents working at once, in the cloud, on the model of your choice.

If you want to build an app without writing code, more guided tools exist: see our piece on no-code app building. For other everyday AI tools, see our AI productivity tools roundup.

Who Is Cursor For?

• Good fit: developers who already hand whole tasks to agents (features, migrations, bug fixes), and teams that want automated code review and agents in Slack or GitHub.
• Start on Hobby: occasional coders, who may be fine on the free plan or a cheaper extension.
• Proceed with care: beginners (understand generated code before accepting it) and companies that need data kept outside the US.

FAQ

Is Cursor free?

Yes. The Hobby plan is free with no credit card, but Agent requests are limited and Cursor doesn’t publish the number. For daily use, you’ll need Pro ($20/month) or higher.

How much is Cursor Pro?

$20 a month, or $16 a month billed yearly, before tax. Pro+ is $60 and Ultra $200 (prices checked Sept. 25, 2026).

Does Cursor train on my code?

Not with Privacy Mode on: neither Cursor nor its model providers train on your code. With Privacy Mode off, Cursor may use your code to improve and train its models.

What happens when I hit my usage limit?

Cursor notifies you in the editor. You can turn on on-demand usage at the same rates or upgrade your plan. Usage resets every billing cycle.

Does Cursor work with VS Code extensions?

Yes. Cursor’s editor is a VS Code fork and keeps its extensions; the newer Agents Window is a separate interface, and you can switch between the two at any time.

Who owns Cursor?

Cursor is built by Anysphere, which SpaceX acquired (announced August 14, 2026). SpaceXAI’s Grok models now sit alongside models from Anthropic, OpenAI and Google.