Cursor review
- Made by
- Anysphere, a SpaceX subsidiary since 14 Aug 2026
- Founded
- 2022 (headquarters: San Francisco)
- Free plan
- Yes (Hobby: limited Agent requests and Tab completions)
- Paid plans from
- $20/month Pro ($16/month billed yearly)
- Team plan
- $40/user/month ($32 billed yearly)
- Platforms
- Windows, macOS and Linux editor, CLI, iOS app
- Models
- Grok 4.7, Composer 2.5, Claude, GPT and Gemini
- Security
- SOC 2 Type II, ISO 27001, ISO 42001, AIUC-1
Cursor is a code editor with AI built into every part of it. It is based on VS Code, so your extensions, themes and shortcuts carry over, and on top it adds predictive tab completion, an agent that plans and edits across many files, cloud agents that work in the background, a command-line tool and Bugbot for pull-request review. You can switch between Claude, GPT, Gemini, Grok and Cursor's own Composer models.
There is a free Hobby plan, and Pro costs $20 a month. Cursor's maker, Anysphere, became part of SpaceX on 14 August 2026, and Cursor now ships SpaceXAI's Grok models first. It ranks #3 in our best AI for coding ranking, behind Claude Code and OpenAI Codex, and it has the best editor experience of any tool we tested.
Cursor is worth it if you spend your day inside a code editor and want AI that helps without taking over. The review-and-accept flow and tab completion are the best we have used, and model choice lets you trade cost for quality task by task. The catch is cost predictability: plans give you a pool of usage, not a fixed number of requests, and heavy agent users often end up on Pro+ ($60) or Ultra ($200).
Score breakdown
Best for
- Developers who live in their editor and want to review every change
- VS Code users who want AI without learning a new tool
- People who want Claude, GPT, Gemini and Grok models in one place
- Teams that want cloud agents, including on machines they manage
Skip it if
- Anyone who needs a fixed, predictable monthly bill for heavy agent use
- JetBrains, Visual Studio or Xcode users who will not change editor
- Companies uneasy about the new SpaceX ownership
- Developers who prefer delegating whole tasks from a terminal
What we like
- Best editor experience: fast tab completion and clear diffs to accept or reject
- Choice of Claude, GPT, Gemini, Grok and Cursor's own models, with automatic routing
- Familiar VS Code base, so extensions and shortcuts carry over
- Cloud agents that run in parallel, including on your own infrastructure
- Unlimited tab completions on every paid plan
Watch out for
- Usage pools make monthly cost hard to predict
- Heavy agent users need Pro+ ($60) or Ultra ($200)
- Very poor Trustpilot score, driven by billing and support complaints
- New SpaceX ownership may raise procurement questions
Key features, explained
Agent
Cursor's agent takes a request in plain English, reads the relevant files, plans the change and edits across the project. It can run terminal commands and tests, then shows every change as a diff you can accept or reject. That review step is Cursor's biggest strength: you stay in control of each edit instead of trusting a finished pull request. Rules, skills, hooks and MCP servers let you shape how it works.
Tab completion
Tab is Cursor's predictive autocomplete. It suggests the next edit, often several lines at once, and you press Tab to accept. Reviewers on Capterra and Product Hunt single it out for accurate multi-line predictions. Tab completions are unlimited on Pro, Pro+ and Ultra; the free Hobby plan has a limited number.
Model choice and two usage pools
You can pick a model per task or let Cursor's Auto setting choose. Usage comes from two monthly pools. The Cursor Models pool covers Grok 4.7, 4.6, 4.5 and Composer 2.5, with much more included usage. The Other Models pool covers Claude, GPT and Gemini at each provider's API price, so choosing Claude Opus 5.5 ($4/$20 per million tokens) drains it faster than Grok 4.7 ($2/$6).
Cloud agents
Cloud agents work on tasks in the background, in parallel, for hours at a time, while you keep coding. Cursor's iOS app lets you follow them away from your desk. Since March 2026, companies can also run them on machines they manage, so code and build artifacts stay inside their own network. Cursor says cloud agents now start 3x faster than before.
Cursor CLI
The CLI brings Cursor's agent to any terminal window, for people who like the Claude Code or Codex style of working, and it suits scripts and remote servers where no editor is open. On the independent Terminal-Bench 2.1 leaderboard, Cursor CLI with Grok 4.5 scored 79.3%, fifth overall, behind Codex CLI with GPT-6 Astra (87.4%) and Claude Code with Claude Fable 5 (83.8%).
Bugbot code review
Bugbot reviews pull requests on GitHub and hunts for real bugs rather than style issues. It is billed per review from your included usage and then from on-demand spend, instead of a fixed seat fee. You can set effort levels and team rules, and send flagged issues straight to Cursor's agents to fix. See how it compares in our best AI code review ranking.
Privacy mode
With Privacy Mode on, Cursor says neither it nor its model providers will train on your code, backed by contracts with those providers. Any user can switch it on in settings, and team admins can enforce it for everyone. Cursor holds a SOC 2 Type II attestation and ISO 27001, ISO 42001 and AIUC-1 certifications, which helps with company security reviews.
Where Cursor ranks
Our scores come from our category rankings, where each tool is scored 0–10 on weighted criteria against its rivals.
| Ranking | Listed as | Position | Score |
|---|---|---|---|
| Best AI for Coding | Cursor | #3 of 13 | 9.0 |
What users say about Cursor
Opinion on Cursor splits sharply by where you look. Developer communities rate it very highly: it averages 5.0 on Product Hunt from 951 reviews and 5.0 on Capterra from a small sample of 4. Reviewers there praise how well it understands a whole project, the VS Code base, fast tab completion and smooth multi-file edits. Trustpilot tells a different story, with a 1.6 TrustScore from 331 reviews as of September 2026, most of them one star. Those complaints are mostly about money and support: surprise on-demand charges, trouble getting refunds, and an AI support agent that is hard to get past. Across all sites, users also mention slowdowns and heavy memory use on large repositories, an agent that can be confidently wrong, and pricing that is hard to understand. Cursor's 2025 switch to usage-based pricing caused a backlash, and it apologised.
Users praise
- Understands the whole project across files
- Fast, accurate tab completion that predicts multi-line edits
- Keeps VS Code extensions, themes and shortcuts
- Easy switching between Claude, GPT, Gemini and Grok
Users complain about
- Unexpected on-demand charges and slow refunds
- Support is mostly an AI agent, with human help hard to reach
- Slow and memory-hungry on large repositories
- Usage-based pricing is hard to understand
| Review site | Rating | Reviews | Checked |
|---|---|---|---|
| Product Hunt | 5/5 | 951 | 2026-09 |
| Capterra | 5/5 | 4 | 2026-09 |
| Trustpilot | 1.6/5 | 331 | 2026-09 |
Ratings are third-party aggregates and change over time. They are context, not part of our score.
Cursor has a free Hobby plan, and paid plans start at $20 a month for Pro ($16 a month billed yearly). Pro+ costs $60 ($48 yearly) with 3x Pro's usage on frontier models, and Ultra costs $200 ($160 yearly) with 20x. Teams costs $40 per user per month ($32 yearly) for a standard seat or $120 ($96 yearly) for a premium seat with 5x the agent limits. Enterprise is custom.
Every plan includes a set amount of model usage split into two pools: a generous one for Cursor's Grok and Composer models, and one for Claude, GPT and Gemini at API prices. When it runs out you pay on demand or upgrade. Cursor itself recommends Pro+ for daily agent users and Ultra for power users. Prices are in US dollars, exclude tax and were checked on 25 September 2026.
| Plan | Price |
|---|---|
| Hobby | Free |
| Pro | $20/month, or $16/month billed yearly ($192/year) |
| Pro+ | $60/month, or $48/month billed yearly ($576/year) |
| Ultra | $200/month, or $160/month billed yearly ($1,920/year) |
| Teams (Standard) | $40/user/month, or $32/user/month billed yearly |
| Teams (Premium) | $120/user/month, or $96/user/month billed yearly |
| Enterprise | Custom |
Full Cursor pricing breakdown, hidden costs and money-saving tips
How Cursor works
You download the Cursor editor for Windows, macOS or Linux. Because it is built on VS Code, most extensions, themes and shortcuts work as before. From there you code as usual: Tab suggests the next edit as you type, a keyboard shortcut asks for a targeted change to selected code, and the agent takes bigger jobs that span many files.
Every plan includes a set amount of model usage each month. Cursor splits it into two pools: one for its own Grok and Composer models, with generous limits, and one for third-party models such as Claude and GPT, charged at the provider's API price. When a pool runs out you can turn on on-demand usage (pay as you go, billed after the fact) or upgrade. Usage resets with your billing cycle and does not roll over. Full details are on our Cursor pricing page.
The SpaceX deal and what it changed
SpaceX's purchase of Anysphere closed on 14 August 2026, valuing Cursor at $60 billion, and Cursor is now a wholly owned SpaceX subsidiary. The visible changes so far:
- Grok models come first. Grok 4.7 arrived on 21 September 2026, and Cursor calls it its most capable model for long-running coding work. Grok models get the largest included allowance on every paid plan.
- Grok Bot, an autonomous agent that signs into your tools and does work for you, is included on paid plans.
- Other models remain. Claude, GPT and Gemini are still available, billed at API prices from the second usage pool.
If your company has rules about vendor ownership or where data goes, check them before rolling Cursor out.
Who should skip Cursor
Cursor is an editor first. If you use JetBrains, Visual Studio or Xcode and do not want to switch, GitHub Copilot works inside those. If you would rather hand a whole ticket to an agent and review the pull request, Claude Code or OpenAI Codex fit better. And if you need to know your bill in advance, a flat plan with hard limits is easier to budget than Cursor's usage pools and on-demand billing. Our Cursor alternatives page compares the options, and Cursor vs GitHub Copilot goes head to head.
- Send routine edits, tests and refactors to Grok 4.7 or Composer 2.5. They draw from the larger Cursor Models pool, so your Claude and GPT allowance lasts for the hard problems.
- Set an on-demand spend limit in your dashboard on day one, then check the usage page weekly. Most bad reviews start with an unexpected on-demand bill.
- Write project rules for your build, test and style commands. The agent follows them on every run, which saves both usage and review time.
- Turn on Privacy Mode before opening client or company code if your team admin has not already enforced it.
- Use cloud agents for well-scoped tickets with a clear test for 'done', and keep interactive agent work for tasks where you want to steer.
Cursor alternatives
Jargon explained
- Tab completion
- Autocomplete that predicts your next edit, often several lines at once. You press Tab to accept it.
- Usage pool
- A monthly allowance of AI use included with your plan. Different models drain it at different speeds, based on their price per token.
- On-demand usage
- Pay-as-you-go billing that keeps the AI running after your included usage runs out. Cursor bills it after the fact.
- Cloud agent
- An AI agent that works on a task on a remote computer in the background, then hands back the result, often as a pull request.
- Privacy Mode
- A Cursor setting that stops Cursor and its model providers from training on your code.
Frequently asked questions
Is Cursor free?
Yes, the Hobby plan is free with no credit card. It includes limited Agent requests, limited Tab completions and access to Cursor's Composer model. Pro, at $20 a month, adds unlimited Tab, frontier models and cloud agents. See Cursor pricing.
Is Cursor better than VS Code with Copilot?
Cursor has the smoother AI editing experience and more model choice; Copilot is cheaper ($10 vs $20) and works in more editors. Cursor ranks above Copilot in our best AI for coding list. See Cursor vs GitHub Copilot.
Who owns Cursor now?
SpaceX. Its acquisition of Cursor's maker, Anysphere, closed on 14 August 2026 at a $60 billion valuation. Cursor now operates as a wholly owned SpaceX subsidiary.
Which AI models does Cursor use?
Grok 4.7, 4.6 and 4.5 from SpaceXAI and Cursor's own Composer 2.5 (in the larger Cursor Models pool), plus third-party models such as Claude Opus 5.5, GPT and Gemini, billed at their API prices.
Does Cursor train on my code?
Not with Privacy Mode on. Cursor says it guarantees that neither it nor its model providers train on your code when Privacy Mode is enabled. Any user can turn it on, and team admins can enforce it.
Can I use my VS Code extensions in Cursor?
Yes, in most cases. Cursor is built on VS Code, so most extensions, themes and keyboard shortcuts carry over.
Sources
Every fact on this page comes from public information. Vendor figures are labelled as vendor claims.
- Cursor pricing (Cursor)
- Plans and usage (Cursor docs) (Cursor)
- Models and pricing (Cursor docs) (Cursor)
- Usage and limits (Cursor help) (Cursor)
- Increased usage for agents (Cursor)
- Cursor blog (SpaceX, Grok 4.7, cloud agents) (Cursor)
- Introducing Grok 4.7 (Cursor)
- Self-hosted cloud agents (Cursor)
- Bugbot (Cursor docs) (Cursor)
- Cursor security (Cursor)
- Cursor for students (Cursor)
- Cursor product site (Cursor)
- Cursor (company) (Wikipedia)
- Cursor reviews (Trustpilot)
- Cursor reviews (Product Hunt)
- Cursor reviews (Capterra)
- Terminal-Bench 2.1 leaderboard (Snorkel AI / Terminal-Bench)
- GitHub Copilot plans (GitHub)
- Claude plans and pricing (Anthropic)
- Codex pricing and plan limits (OpenAI)
- openai/codex repository (GitHub)
- Google Antigravity pricing (Google)
- Devin pricing (Cognition)
- Windsurf pricing (redirects to Devin) (Windsurf / Cognition)
- Cline pricing (Cline)