Cursor is worth paying for in 2026 if you want tab completion, a coding agent and models from several vendors in one editor. Paid plans run from $8 to $200 a month, but third-party models bill at API price, so Cursor's own docs put daily agent users at $60 to $100 a month. The gaps are an unpredictable bill, Open VSX extensions only, and privacy that depends on a setting. This review is built from Cursor's documentation, pricing page and changelog.
Cursor AI review, short version: Cursor is worth paying for in 2026 if you want tab completion, a coding agent and models from several vendors in one editor, and it is the wrong buy if you need a bill that never moves. On 13 October 2026 its pricing page listed Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200, with third-party models billed separately at each model's API price.
Prices and features checked October 2026 against Cursor's pricing page (plan names and prices as read on 13 October 2026), its docs on models and pricing, Tab, Agent, modes, rules, Cloud Agents, Bugbot, extensions and spend limits, the data-use page and the changelog.
A note on method, because "review" can mean anything. This one is a documentation review: every claim below comes from a page Cursor publishes, read this month. We ran no benchmark and we print no speed scores, so what you get is what the product does, what it costs and what the fine print says. For the wider field of AI coding tools, start with our Claude Code complete guide, which is the hub for this topic.
Cursor AI review: the verdict in one table
Cursor does more than its pricing page can summarise, so here is the scorecard first. Each row pairs what the docs say you get with the catch written in those same docs.
| Area | What you get | The catch |
|---|---|---|
| Tab completion | Multi-line edits, jump-in-file and cross-file suggestions; unlimited on paid individual plans, per the docs | The pricing page lists no tab allowance for any plan |
| Agent | Searches, edits files, runs shell commands and drives a browser, with unlimited tool calls and checkpoints | Tool calls are unlimited; the tokens they burn are not |
| Plan Mode | Researches the codebase, asks questions and writes an editable plan before any code | Plans save to your home directory until you move them into the workspace |
| Models | OpenAI, Anthropic, Google and others, plus Cursor's own Grok and Composer models | Third-party models bill at API price, on top of the plan price |
| Cloud Agents | Isolated cloud VMs started from desktop, web, iOS, Slack, GitHub, Linear or the API | Charged at API pricing for the selected model |
| Bugbot | Pull request review on GitHub, GitLab, Bitbucket and Azure DevOps | Usage-based, and the docs page prints no dollar rate |
| Privacy | Privacy Mode: no training on your code, zero-data-retention agreements with providers | With it off, code and prompts may be stored and used for training |
| Extensions | The Open VSX registry, plus Cursor-built replacements for some popular extensions | Microsoft Marketplace-only extensions are not all available |
Read the right-hand column as the review. None of those catches is hidden; each one sits in Cursor's own documentation. They are simply not on the pricing page, which is where most people stop reading.
What Cursor is in 2026
Cursor is a desktop code editor with AI in every layer, and its docs now describe it as "a coding agent for building ambitious software". The editor is the base. Around it sit a command-line agent, cloud agents that keep working with your laptop closed, a pull request reviewer and, since September, deploy monitoring.
- 01Tab
Completions as you type, including edits in other files.
- 02Plan Mode
Reads the codebase and writes a plan you can edit.
- 03Agent
Edits files, runs commands and checks the result in a browser.
- 04Cloud Agents
The same job in a cloud VM, started from Slack, the web or your phone.
- 05Bugbot
Reviews the pull request and proposes fixes.
- 06Rollouts
Watches the deploy and reports health per environment.
You do not need all six. Most solo builders live in the first three: Tab while typing, Plan Mode before anything large and Agent to carry it out. The last three matter once work leaves your machine, and they are where the extra spend comes from.
The controls are few. Agent opens in the side pane with Cmd+I, Shift+Tab in the chat input rotates to Plan Mode, and the CLI is a separate install whose command is agent, with agent -p for scripts and CI. If you want a hands-on walkthrough of the editor itself, our Cursor tutorial covers setup and a first build.
What does Cursor cost in 2026?
Cursor's pricing page lists one free plan, four paid individual plans, Teams and Enterprise. The plan price buys features plus an amount of included usage. It does not cap what you can spend.
Source: cursor.com/pricing, read 13 October 2026. Prices exclude tax.
| Plan | Price (read 13 October 2026) | What the plan lists |
|---|---|---|
| Free | $0 | Limited agent requests and limited chat requests |
| Plus | $8 a month | The introductory plan: a usage pool shared across products, Grok models only |
| Premium | $30 a month | 4x the usage of Plus, cloud agents and automations, code hosting, AI code reviews, third-party models at API list price |
| Super | $100 a month | 15x the usage of Plus, priority model access at peak times |
| Ultra | $200 a month | 50x the usage of Plus, earliest access to products and models |
| Teams | From $30 a user a month | Central billing, a team marketplace for rules, skills and plugins, SAML/OIDC SSO, usage analytics, team-wide privacy settings |
| Enterprise | Custom | Pooled usage, invoice billing, SCIM, audit logs, repository, model and MCP access controls |
Those are the names and prices on the pricing page. Cursor's docs and help centre still listed the earlier line-up that day: a free Hobby plan, Pro at $20, Pro Plus at $60 and Ultra at $200, with Teams seats at $40 and $120. The page also says third-party model usage is not included in the monthly price and has to be bought separately.
Under the earlier plan names, the docs split a paid individual plan's usage into two pools that reset with your billing cycle. The Cursor Models pool covers Grok 4.7, 4.6 and 4.5 and Composer 2.5, and Cursor says it carries significantly more included usage. The Other Models pool covers every third-party model, charged at that model's API price.
That split is the whole pricing story. The model you pick decides how fast the allowance drains, and the spread between models is wide.
The first two draw on the Cursor Models pool; the rest draw on Other Models. Source: cursor.com/docs/models-and-pricing, checked October 2026.
Composer 2.5 output costs a twentieth of Fable 5.1 output. An afternoon of agent runs on a frontier model moves the meter far faster than the same afternoon on Cursor's own models. When the allowance is gone you either upgrade or turn on on-demand usage, which Cursor bills monthly at the same rates. The docs add that requests are never downgraded in quality or speed.
Cursor publishes its own guidance on where people land: daily Agent users typically total $60 to $100 a month, and power users running several agents or automations often pass $200. Treat those as the realistic prices of the product. The plan price is the entry fee.
Why older guides quote 500 requests
If you have read that Pro includes 500 fast requests, that was the old model, and this page quoted it too until this update. Cursor's July 2025 post, Clarifying our pricing, said Pro had moved from request limits to included usage at API pricing, admitted the rollout was unclear and offered refunds for unexpected charges. Today's docs mention the old system only as "legacy request-based plans", and Max Mode is now available only on those.
Cursor's strengths
Six things in the documentation justify the subscription. They are listed in the order most people will feel them.
- Tab is still the reason to choose an editor. It suggests multi-line edits, adds missing imports, jumps to the next place you will need to edit and offers changes in other files through a portal at the bottom of the window. Cursor's docs list unlimited completions on paid individual plans.
- The agent has a real undo. Agent snapshots your files before significant changes, and you can preview or restore a checkpoint from the chat timeline. You can also queue a follow-up with Enter or steer the current turn without stopping it.
- Plan Mode slows the agent down in the right place. It researches the codebase and asks clarifying questions, then gives you a plan to edit before a line of code is written. For anything touching more than one file, that review step is cheaper than untangling a wrong build.
- A model menu, not a single vendor. You can run cheap work on Composer 2.5 and switch to a frontier model from Anthropic, OpenAI or Google for the hard task, inside the same chat.
- Rules that live in the repo. Project rules in
.cursor/rulesare version-controlled and can apply always, by file pattern, by description or only when mentioned. A plainAGENTS.mdworks too. - It reaches past your desk. Cloud Agents start from the web, iOS, Slack, a GitHub comment or Linear, and since 6 October 2026 the iOS app can view and reply to agents running locally on your computer.
Cursor's gaps: what the docs admit
The gaps are mostly about money, data and portability, not about what the agent can do. Here they are next to the strengths, then one at a time.
- The allowance has no published dollar size
- Third-party models bill at API price
- Teams pay a token fee on third-party models
- Privacy depends on a setting you must check
- Extensions come from Open VSX only
- Cloud Agents and Bugbot are metered separately
- Unlimited tab completions on paid individual plans
- Agent with checkpoints, queueing and steering
- Plan Mode before large changes
- Models from several vendors in one menu
- Repo-level rules and AGENTS.md support
- Agents on web, iOS, Slack, GitHub and Linear
- The bill is hard to predict. Neither the pricing page nor the models doc prints the dollar size of the included usage; the pricing page sizes each plan only as a multiple of Plus. You learn your real number from the usage dashboard after a few weeks.
- Teams pay an extra token fee. On Teams and Enterprise, third-party model requests carry a Cursor Token Rate of $0.25 per million tokens on top of the model's price. Cursor's own Grok and Composer models are exempt.
- Privacy is a setting, not an assumption. With Privacy Mode off, the data-use page says Cursor may store and use codebase data, prompts, editor actions and code snippets to train its models. The page does not say which state is the default, so open the setting and look. Fable 5.1 also needs data-retention approval even with Privacy Mode on.
- Not every extension comes with you. Cursor uses the Open VSX registry. Most popular extensions are there, and Cursor ships audited replacements for some that are not, but a Microsoft Marketplace-only extension can be missing.
- Cloud Agents are a second meter. The pricing page lists them from Premium up, and they are charged at API pricing for the model you select. User-level hooks from
~/.cursor/hooks.jsondo not run there. - Checkpoints are not version control. They are stored locally, separate from Git, and restoring one reverts files only, not the conversation. Commit before a long agent run.
- Small format traps. Project rules must use the
.mdcextension; a plain.mdfile in.cursor/rulesis ignored. The rules page no longer mentions the old.cursorrulesfile that many tutorials still recommend. - New features arrive unevenly. The docs mark
/goaland steering in the Agents Window as rolling out, so a feature in the changelog may not be in your build yet.
Is Cursor AI worth it? Who it is for
Cursor is worth it when two things are true: you want completions and an agent in the same window, and you value switching between model vendors. If only one is true, a cheaper or flatter option usually fits better.
Good fit:
- Solo builders shipping a web product who want one tool for typing, planning and agent work.
- Developers coming from VS Code whose extensions are on Open VSX.
- Small teams that want shared rules, SSO and Bugbot reviewing pull requests under one bill.
Poor fit:
- Anyone who needs a fixed cost. Set a spend limit, or accept that on-demand usage stops when you hit it.
- Workflows tied to a Microsoft-only extension. Check before you migrate, not after.
- Heavy users of one vendor's models. If nearly all your agent work runs on Claude, read our Claude Code vs Cursor comparison and price a Claude plan against your Cursor dashboard.
Cursor review 2026: what changed this year
A changelog tells you where a product is heading. Cursor's last ten weeks of releases are almost all about agents that run somewhere other than your editor.
- Oct 6Remote control for local agents
View and reply to agents on your computer from the iOS app.
- Sep 23Rollouts and Security Review
Deploy health per environment, and one security review comment per pull request.
- Sep 10Cursor Projects
Long-running work on cloud agents, with a coordinator agent and shared context.
- Sep 2Self-hosted machines
Keep tool execution inside your own network.
- Aug 27Start from scratch, without a repo
Cloud Agents no longer need a connected GitHub repository to begin.
- Aug 19/goal, custom modes and subagent machines
A long-lived objective for the agent, and an isolated environment per subagent.
- Aug 17Origin code hosting
Repositories hosted by Cursor, with pull requests and GitHub sync.
- Jul 29Cursor on iPad
A review surface with comments, checks and approvals.
Source: cursor.com/changelog, checked October 2026.
Read together, the releases move Cursor from an editor with an assistant towards a platform that hosts agents, repositories and deploy checks. That is good news if you want to hand off long tasks. If you only want a fast editor with completions, note that most of this year's work, and most of the usage-billed features, sit outside it.
How to trial Cursor without overspending
Give it two weeks and a spending cap before you judge it. The order below keeps the test cheap and the result honest.
- 1Start on Free
Learn Agent, Plan Mode and the shortcuts on a small repo.
- 2Check Privacy Mode
Open the setting before any client or employer code goes in.
- 3Write one rule file
An AGENTS.md or a .mdc rule with your stack, test command and conventions.
- 4Move to a paid plan, monthly
Not yearly yet. You are buying a month of data about your own usage.
- 5Cap on-demand spend
Leave on-demand off, or switch it on with a spend limit in the dashboard.
- 6Default to Cursor models
Use Composer or Grok for routine work and pick a frontier model per task.
- 7Read both pools weekly
The usage dashboard shows Cursor Models and Other Models separately.
- 8Decide at day 14
Stay on that plan, move up a tier, or move heavy agent work to a flat plan.
- Privacy Mode is in the state you expect
- On-demand usage is off, or has a spend limit
- Your must-have extensions exist on Open VSX
- Project rules use the .mdc extension, or live in AGENTS.md
- The default model is one from the Cursor Models pool
- Cloud Agents have their own spend limit before the first run
- Git is clean and committed before any long agent run
The editor is the easy part. Knowing what to build with it, then getting it deployed and billed, is where projects stall. Our AI SaaS Builder program includes a lesson on building with Cursor and sets up a Cursor, Supabase and Vercel environment for its MVP build, then carries the product through launch and Stripe billing.
Alternatives if Cursor is not the fit
The two questions that sorted Cursor sort every tool in this field: editor or agent, and one vendor or many. Claude Code is the agent-first, single-vendor answer, and the Claude Code vs Cursor comparison sets the two side by side with current prices.
If cost is the blocker, our Claude pricing guide explains the flat-plan side of the trade, and how to use Claude Code for free covers what a zero budget realistically gets you. Whichever way you go, run the two-week trial above before you commit to an annual plan.
Cursor AI review: FAQ
Is Cursor AI worth it in 2026?
Cursor is worth it if you write code most days and want tab completion, an agent and a choice of model vendors in one editor. Its pricing page listed Plus at $8 a month and Premium at $30 when read on 13 October 2026. It is harder to justify if you need a fixed bill: Cursor's own docs say daily Agent users typically total $60 to $100 a month and power users often pass $200, because third-party models bill at API price.
How much does Cursor cost?
Cursor's pricing page, read on 13 October 2026, lists a Free plan, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200. Teams starts at $30 per user a month and Enterprise is custom. Plus runs Grok models only. From Premium up you can pick third-party models, which are billed separately by usage at API list price. Prices exclude tax. Cursor's docs and help centre still listed the earlier Hobby, Pro, Pro Plus and Ultra plans that day.
Is Cursor free to use?
Yes, within limits. The Free plan includes limited agent requests and limited chat requests. Cloud agents, automations, AI code reviews and third-party models are listed from Premium, at $30 a month, and the $8 Plus plan runs Grok models only. Free is enough to learn the editor, not to work in it all day. Read from Cursor's pricing page on 13 October 2026.
Does Cursor train on my code?
It depends on Privacy Mode. Cursor's data-use page says that with Privacy Mode on, customer data is not used for training by Cursor, and Cursor keeps zero-data-retention agreements with its model providers. With Privacy Mode off, Cursor may store and use codebase data, prompts, editor actions and code snippets to improve its features and train its models. Check the setting before you open client code.
Do VS Code extensions work in Cursor?
Most do, but not all. Cursor's docs say it uses the Open VSX registry for third-party extensions, not Microsoft's marketplace. Most popular extensions are listed on Open VSX, and Cursor publishes its own audited replacements for some widely used ones that are not. An extension that exists only on the Microsoft Marketplace may be unavailable, so check your must-have list before you switch.
What happened to Cursor's 500 requests a month?
Cursor replaced it. In a July 2025 post titled Clarifying our pricing, Cursor said Pro had moved from 500 requests a month to included usage billed at API pricing, and offered refunds for unexpected charges during the rollout. The current docs describe two monthly usage pools and mention the old model only as legacy request-based plans. Any guide that still quotes 500 fast requests is out of date.
Should I use Cursor or Claude Code?
Use Cursor if you want an editor with tab completion and models from several vendors. Use Claude Code if you want Anthropic's agent in the terminal on a flat Claude subscription. They are not exclusive: Anthropic's VS Code extension docs include an install link for Cursor, so the Claude Code panel can run inside the Cursor editor on your Claude plan while Cursor's Tab keeps working.
Free · no card needed
Start with the free Starter Kit.
A unit economics calculator from AI SaaS Builder, plus four more working files.
Chosen your editor? Ship a product with it.
All Access includes AI SaaS Builder, which has a lesson on building with Cursor and takes one product from Supabase schema to a Claude API feature, a Vercel deploy and Stripe billing, alongside the other three programs, live coaching and the private community.
Not sure Cursor fits?
Ask in the free Discord before you pay for a plan. It is where builders compare notes on AI coding tools, plans and usage bills.