Cursor AI: Pricing, Models & Comparison (2026)

👁 119 reads

Cursor is a code editor from Anysphere built on VS Code, where an AI agent reads your entire project, edits multiple files, runs commands, and can work in the cloud when you close your laptop. The Hobby plan is free, Pro costs $20 per month, and since August 2026, Cursor has been owned by SpaceX. Below, we’ve gathered everything you need to know about what it can do today, which models it offers, how much it costs, and what to compare it with.

What is Cursor

Cursor is a standalone application for Windows, macOS, and Linux, not a plugin. It started as a fork of VS Code, so it looks familiar, supports the same extensions and shortcuts. The difference is that the language model is built into the editor: it indexes your repository, sees the project structure, and can open files, make changes, and run terminal commands on its own.

Cursor 3 shifted the focus from writing code to managing agents. Instead of the old Composer panel, there’s now an Agents Window - a window where you launch multiple agents in parallel, locally, in the cloud, or on a remote machine via SSH. The classic editor view still works alongside it. Beyond the desktop app, there’s a web version (cursor.com/agents), an iOS and iPad app, and the ability to launch agents from Slack, GitHub, and Linear.

The company was founded in 2022 by four MIT graduates. We covered the acquisition story and its consequences in SpaceX buys Cursor for $60 billion and OpenAI cuts Cursor off from its models. Since August, Cursor also has its own repository hosting - Origin, a GitHub rival, in beta for paid plans.

Installation and Getting Started

1. Download the installer from cursor.com/download (Windows x64 and ARM64, macOS, Linux) and log in with a Cursor account - the Hobby plan doesn’t require a card.
2. On first launch, choose to import from VS Code. Cursor transfers settings, theme, shortcuts, and extensions with one click.
3. Open a project folder. Cursor will start indexing it so the agent can search through the code.
4. Learn three entry points: Ctrl+K (Cmd+K on Mac) edits a selected fragment, Ctrl+L opens the chat panel with the agent, and from the command palette (Ctrl+Shift+P) you open the Agents Window.
5. Set the command execution mode in Settings → Cursor Settings → Agents → Run Mode. Since version 3.6, the default recommended mode is Auto-review: safe commands run immediately, the rest go to a classifier or ask for permission. Cursor itself notes this is for convenience, not a security boundary.

Responses in Other Languages via Rules

Cursor’s interface is in English and we didn’t find an option for other language versions in the documentation. However, you can chat with the agent in other languages, and to avoid reminding it in every message, you can add a rule. There are two places:

• User Rules - globally for all projects, in Customize → Rules. For example, you could enter: “Always respond in [your language]. Write variable names, code comments, and commit descriptions in English.” Note: User Rules work in the agent but not in Ctrl+K editing.
• Project Rules - a file in the .cursor/rules directory (.mdc format with description, globs, and alwaysApply fields) or a simple AGENTS.md in the main repository directory. This version goes into version control, so the whole team gets the same rules.

You should also add project conventions to the rules: framework, how to run tests, what not to touch. Without them, the agent guesses. We detailed the complete first session step by step in the guide Cursor step by step: first hour with the AI editor.

Key Features

• Tab - suggestions on the fly, including multi-line and spanning multiple places in a file at once, e.g., when renaming. Unlimited in paid plans.
• Agent - you describe a task, and the agent reads files, edits, runs tests, and shows diffs for approval. There are also Ask mode (questions only, no changes), Plan mode (plan first, then execution), and custom modes. Since September, you can add commands while the agent is working without interrupting it.
• Cloud Agents (formerly Background Agents) - agents running in Cursor’s isolated virtual machines with a cloned repository, dependencies, and network access. You launch them from the editor, browser, phone, Slack, GitHub, Linear, or via API. They require a paid plan and are billed at API prices for the chosen model. Since September 2, you can run them on your own machines (self-hosted), so code doesn’t have to leave your corporate network.
• Projects (beta since September 10) - a coordinating agent divides larger work, like a migration, into tasks for sub-agents. It can also monitor a Slack channel or run on a schedule.
• Bugbot - automatic pull request review on GitHub, GitLab, Bitbucket, and Azure DevOps. It leaves comments with explanations and proposed fixes, and the Autofix option launches a cloud agent that pushes the fix itself. Bugbot rules are saved in .cursor/BUGBOT.md. In individual plans, it’s billed by usage; in Teams, it’s included. For Teams and Enterprise, Security Review and Rollouts bots were added on September 23.
• Rules - User, Project, and Team Rules described above. Priority order: Team → Project → User.
• MCP, skills, hooks, and Marketplace - Cursor supports MCP servers, so the agent can reach databases, Figma, Jira, or documentation. Teams can share MCP, rules, and plugins in a common marketplace.

We described six specific tasks where this translates into saved time in Cursor in practice.

Models in Cursor

Cursor isn’t tied to a single provider. The model selection has dozens of options, and Auto mode automatically selects a model in one of three settings: Cost, Balance, and Intelligence. Auto isn’t free - each query costs as much as the model it went to.

The list also includes Claude Fable 5 and 5.1, Kimi K3, GLM 5.3, and older versions of Claude, GPT, and Gemini. Three things worth knowing:

• GPT-6 is not in Cursor. OpenAI refused to provide new models, including GPT-6 Astra, after the SpaceX acquisition. According to Cursor’s CEO, OpenAI models handled about 5% of traffic. If you need GPT-6, the natural choice is Codex.
• Cursor’s models have a separate, larger pool. Composer 2.5 and Grok 4.x count toward the “Cursor Models” pool, while Claude, Gemini, and GPT count toward “Other Models.” With the same subscription, you can do more on Cursor’s own models.
• Watch your default model. Cursor forum users reported the app automatically switching to Grok. Setting Settings → Agents → Conversation → Default Model to “Last Used” helps.

We covered when to switch from Grok to a more expensive model in Grok 4.7 in Cursor, and we write about Sonnet 5.5 itself in Claude Sonnet 5.5: launch and pricing.

How Much Does Cursor Cost

We checked the pricing on October 3, 2026, at cursor.com/pricing and in the documentation. Prices are in dollars, excluding tax.

How billing works:

• Each plan includes an amount of usage in two pools that renew monthly. Usage is counted at the model’s API price, so the same subscription goes much further with Composer than with Opus.
• After exhausting the pool, you can enable on-demand payment for additional usage at the same rates, billed in arrears, or upgrade to a higher plan. It’s worth setting a spending limit right away.
• In Teams and Enterprise, there’s an additional Cursor fee of $0.25 per million tokens for queries to external models.
• Official pages no longer list pool amounts in dollars. External listings mention $20 of external model usage in Pro, $70 in Pro+, and $400 in Ultra - we treat this as unofficial. The same applies to the annual payment discount: the toggle is on the page, and listings mention about $16 monthly for Pro.

We cover details, including how to choose a plan for your usage, in a separate article How much does Cursor cost. We describe exactly what the free plan offers in Cursor for free, and 10 alternatives with prices in the comparison Cursor alternatives.

Cursor for Free - What Hobby Offers

Hobby is enough to test whether working with an agent in the editor suits you. You get the full editor, VS Code import, a limited number of agent queries, and access to the Composer model. What you don’t get:

• Cloud Agents - they require a paid plan.
• A pool for external models at the scale of Pro, so work on Claude or Gemini runs out quickly.
• An explicit limit - Cursor only mentions “limited” queries and doesn’t give a number. We’re not giving one either, because old values from 2025 are outdated.

Our advice: on Hobby, do one real task on your own project, like fixing a specific bug, and only then decide about Pro.

Cursor vs Claude Code vs Codex vs Antigravity vs Copilot

In short: Cursor and Copilot are for people who write code themselves and want help in the editor, while Claude Code and Codex are for those who more often hand entire tasks to an agent. The boundary is blurring, because Cursor 3 is now primarily a tool for managing agents. We detailed the differences between two terminal agents in Codex or Claude Code, and a broader overview is in the guide AI for programming and the comparison of code agents. If you don’t code, first read Cursor won’t help you if you don’t know what to ask and vibe coding step by step.

Privacy and Privacy Mode

• Privacy Mode is available in every plan, including free. When enabled, Cursor guarantees that neither it nor model providers train on your code, and it has zero-retention agreements with providers.
• Exception: providers may run abuse classifiers, and queries that trigger them may be stored for investigation purposes.
• Without Privacy Mode, Cursor reserves the right to use code, queries, and editor actions to improve and train its models. That’s why we always enable it for company code.
• In Teams, the administrator enforces Privacy Mode for the entire team. Cursor has SOC 2 Type II attestation.
• Cloud Agents clone the repository to Cursor’s machines. If code can’t leave the company, you’re left with self-hosted machines or local work.

Frequently Asked Questions

Cursor AI - what is it?

Cursor is a code editor with a built-in AI agent, based on VS Code and developed by Anysphere, now part of SpaceXAI. The agent reads your entire project, edits multiple files, runs commands, and can work in the cloud.

How much does Cursor cost?

Hobby is free, Pro costs $20 per month, Pro+ $60, Ultra $200, and Teams $40 (Standard) or $120 (Premium) per person. After exhausting the included pool, you pay for additional usage according to model API prices.

Is Cursor free?

Yes, the Hobby plan is free and doesn’t require a card. It has a limited number of agent queries and access to the Composer model, but doesn’t include cloud agents. For daily work, you need the Pro plan.

Does Cursor work in other languages?

The interface is in English, but you can chat with the agent in other languages. To have it always respond in your language, add a rule in Customize → Rules like “Always respond in [language]” or save it in an AGENTS.md file in the project.

Cursor or Claude Code - which to choose?

Cursor if you want to work in an editor, see changes in real time, and switch between models from different companies. Claude Code if you prefer to assign large tasks in the terminal and work only with Claude models. Both start at $20 per month.

What are alternatives to Cursor?

The closest are GitHub Copilot (plugin, from $10), Windsurf, now Devin Desktop (from $20), and Antigravity from Google (free start). If you prefer an agent instead of an editor, choose Claude Code or Codex.

Does Cursor have GPT-6?

No. OpenAI is not providing Cursor with new models after the SpaceX acquisition and announced it will discontinue its models on November 12, 2026. Until then, GPT-5.x is available, including GPT-5.6 Sol, Terra, and Luna.

Does Cursor train on my code?

No, if you enable Privacy Mode - then neither Cursor nor model providers use the code for training. Without this mode, Cursor may use data to improve its models.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources and Verification Date

We verified all data on October 3, 2026. Prices and models in Cursor change every few weeks, so compare them with the official pricing before purchasing.

• Cursor - pricing
• Cursor - models and billing
• Cursor - Teams pricing
• Cursor - changelog
• Cursor - Cloud Agents
• Cursor - Bugbot
• Cursor - Rules
• Cursor - data use and Privacy Mode
• Cursor - security
• Cursor - Composer 2.5
• Cursor - Grok 4.7
• cursorhistory.com - version history
• OpenAI - decision on Cursor
• CNBC - OpenAI ends Cursor’s model access
• Anthropic - Claude pricing
• OpenAI - Codex pricing
• Google - Antigravity pricing
• GitHub - Copilot plans