Claude Code, Codex, Gemini CLI, Cursor and GitHub Copilot are all AI coding agents: tools that read your code, change files and run commands for you. The difference is where they live and who makes them. Claude Code (Anthropic) and Codex (OpenAI) run in the terminal, in editors and in the cloud. Cursor is an AI code editor. Copilot is built into GitHub. And Gemini CLI has mostly been replaced: since 18 June 2026, Google points individual users to Antigravity CLI instead.
- Same job, different homes: a terminal, an editor, a web page or GitHub itself.
- Gemini CLI changed: since 18 June 2026, most individuals use Antigravity CLI instead. Paid organisations keep Gemini CLI.
- Free way in: Codex, Cursor, Copilot and Antigravity CLI, yes. Claude Code needs a paid plan or a Console account.
- No winner here: “which to pick” is this site’s reading.
A chatbot that writes code is like a builder who hands you drawings. A coding agent is a builder who walks into your house, picks up the tools and starts the job. So ask where it works, what it can touch, and who checks the result.
Claude Code vs Codex vs Gemini CLI vs Cursor vs Copilot at a glance
The plain point: the five overlap a lot. The table shows where each runs and whether there is a free way in; it does not rank them.
| Tool | Made by, and what it is | Where you use it | Free way in? |
|---|---|---|---|
| Claude Code | Maker: Anthropic. An agentic coding tool that reads your code, edits files and runs commands. | Where: terminal, VS Code, JetBrains, a desktop app and the web. | Free: no. Paid Claude plans or an Anthropic Console account. |
| Codex | Maker: OpenAI. A coding agent with an open-source command-line tool. | Where: the ChatGPT desktop app, an editor extension, the terminal and the cloud. | Free: yes, in every ChatGPT plan. |
| Gemini CLI / Antigravity CLI | Maker: Google. Antigravity CLI replaced Gemini CLI for most people in June 2026. | Where: the terminal, remote environments and SSH sessions. | Free: Antigravity CLI, yes, with weekly limits. Gemini CLI needs a paid licence or API key. |
| Cursor | Maker: Cursor. An AI code editor with an agent built in. | Where: a desktop editor (Mac, Windows, Linux), a terminal tool, the web, an iPhone app and Slack. | Free: yes, with limited agent requests. |
| GitHub Copilot | Maker: GitHub. An AI assistant built into GitHub and the major editors. | Where: editors (VS Code, Visual Studio, JetBrains, Eclipse, Xcode), the terminal and GitHub.com. | Free: yes, Copilot Free. |
Where each line comes from: the sections below quote each maker’s own page. Plans change, and this site lists no prices: each maker’s plans page, linked in the sources, is the current word.
What is each tool, in one line?
The plain point: each maker’s own description shows what it puts first.
Claude Code. Anthropic calls it “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools.”1
Codex. OpenAI’s own project page says “Codex CLI is a coding agent from OpenAI that runs locally on your computer,” and its code is published under the Apache-2.0 licence, so anyone can read it.2 Beyond the terminal, Codex Cloud takes “coding tasks to work through in the cloud, from investigating bugs to building features.”3
Gemini CLI and Antigravity CLI. Google describes Antigravity CLI as “a keyboard-first terminal client that runs Antigravity workflows in local shells, remote cloud environments and SSH sessions.”4
Cursor. Cursor’s documentation opens with “Cursor is a coding agent for building ambitious software.”5 In practice it is an editor you install, with an Agent that “can complete complex coding tasks independently, run terminal commands, and edit code.”6
GitHub Copilot. GitHub calls it “an AI assistant that helps you write, understand, and ship software.”7 Its agent side is the Copilot cloud agent, which “can research a repository, plan changes, and implement them in the background.”8
Claude Code and Codex reach from the terminal to editors and the cloud, Cursor is an editor, Copilot lives inside GitHub, and Google’s terminal tool changed in 2026.
What happened to Gemini CLI?
The plain point: for most individual users it stopped working on 18 June 2026, and Antigravity CLI is Google’s replacement.
Google announced on 19 May 2026 that “On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra,” and the same applied to people using it free through Gemini Code Assist for individuals.9 Organisations with a Standard or Enterprise licence were told “your access remains unchanged,” and Gemini CLI stays reachable through paid API keys.9
What carried over: Google says Antigravity CLI keeps “Agent Skills, Hooks, Subagents, and Extensions,” with extensions now called plugins, but warned there would not be full feature parity at the start.9 Google’s migration guide adds that the new tool still reads your GEMINI.md and AGENTS.md rule files, which are plain text files of standing instructions for the agent.10 The full story is on Gemini CLI is gone: moving to Antigravity CLI.
Terminal, editor or cloud: where does each one work?
The plain point: all five reach beyond a single place. The real choice is between working on your own computer and handing a task to the cloud.
On your computer. Claude Code installs on macOS, Linux and WSL, and in Windows PowerShell or CMD; it also runs inside VS Code and JetBrains editors.1 Codex works in VS Code and editors built on it, including Cursor and Windsurf, while “Xcode and JetBrains IDEs provide their own integrations.”11 Codex is also built into the ChatGPT desktop app, which OpenAI says “is available for macOS, Windows, and Linux.”12 Cursor installs on Mac, Windows and Linux, and its separate terminal tool, Cursor CLI, “lets you interact with AI agents directly from your terminal.”13 Copilot’s agent mode works in VS Code, Visual Studio, JetBrains, Eclipse and Xcode, and its terminal tool runs on Linux, macOS and Windows.14
In the cloud. Here the agent works on a copy of your project on the company’s computers while you do something else. Cursor’s cloud agents “run in isolated VMs in the cloud,” meaning separate virtual computers, and can be started from the web, the iPhone app, Slack or a GitHub comment.15 Copilot’s cloud agent edits files and runs tests “in an ephemeral cloud development environment,” started from an issue, a pull request or a chat on GitHub.com.8 Claude Code on the web runs long tasks with no local setup, and Codex Cloud gives each task its own workspace.1
On your computer, the agent works on your real files. In the cloud, it works on a copy and hands back a change for you to review. Most of the five now offer both.
What do all five have in common?
The plain point: they are converging: the same instruction files now work in several of them.
Skills. A skill is a folder of instructions an agent loads only when a task needs it (see AI skills explained). The format started at Anthropic and was released as an open standard.16 OpenAI says skills work in Codex, GitHub says the same for Copilot, Cursor finds skill folders automatically, and Antigravity CLI lists custom SKILL.md skills.17 Claude Code uses skills to “package repeatable workflows your team can share.”1
Standing instructions. Claude Code reads a CLAUDE.md file at the start of every session, and “can read” an AGENTS.md written for other coding agents instead.1 Antigravity CLI reads AGENTS.md too (source above). In this site’s reading, that makes one plain instruction file the cheapest way to keep your rules when you switch tools.
Models. Some let you choose whose AI model does the work. Cursor offers “its own models” alongside models from OpenAI, Anthropic and Google.18 Copilot’s model list includes OpenAI, Anthropic, Google, Microsoft, Moonshot AI and xAI, and “the models available to you depend on your Copilot plan.”19 Antigravity lists Gemini models alongside Claude and GPT-OSS models.20 Lists change; check the live page.
Is it safe to let a coding agent run commands?
The plain point: it is as safe as what you let it touch and how carefully you check its work. That is true of all five.
All five can run commands, not just suggest them. On your own computer, a command runs with your files and accounts. In this site’s reading, cloud versions are a gentler start: the agent works on a copy and returns a change you approve. Either way, set what the agent may do without asking (see agent permissions) and read the change itself, not the agent’s summary of it.
Unchecked AI code carries real risk. In one well-known study of GitHub Copilot’s suggestions for security-sensitive tasks, the researchers wrote: “Of these, we found approximately 40% to be vulnerable.”21 It tested a 2021 Copilot: read it as a reason to check code, not a score for today’s tools. Vibe coding explains the difference between checking AI code and shipping it unread.
Pick the tool second. First decide what it may change on its own, and who reads the result before it goes live.
Which one should you pick?
The plain point: start with the accounts you already have. This is this site’s reading, not a ranking; no source above names a best tool.
- You already pay for ChatGPT, or use its free plan: try Codex first, since OpenAI lists it in every ChatGPT plan (source below).
- Your team’s work lives in GitHub issues and pull requests: Copilot’s cloud agent starts from those directly.
- You want one editor with AI built in, and a choice of models: Cursor.
- You like the terminal and have a paid Claude plan: Claude Code.
- You have a Google account and no subscription: Antigravity CLI.
- You want an app without touching code: a browser app builder may suit you better; see Lovable vs Bolt vs v0 vs Replit.
Free plans in brief: OpenAI says “ChatGPT Work and Codex are included in your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan.”22 Cursor’s Free plan offers “Limited agent requests.”23 GitHub says: “Use Copilot Free to explore core features with no paid plan required.”7 Antigravity gives users not on AI Pro or Ultra a weekly allowance, and every plan includes “Access to all product features, such as scheduled tasks and the CLI.”24 Anthropic’s plans page says “Claude Code is included in all paid plans,” and marks it “No” for Free.25
Questions people ask
Is Cursor better than Claude Code?
Neither maker’s documentation claims to beat the other. The practical difference is the home: Cursor is an editor you work inside, while Claude Code runs in the terminal and also inside editors, including Cursor itself, through its extension.1
Does Gemini CLI still work?
Only for some people. Google stopped serving it for AI Pro, AI Ultra and free individual users on 18 June 2026; organisations with a Standard or Enterprise licence, and paid API keys, keep it (source above). Everyone else is pointed to Antigravity CLI.
Do I need to know how to code to use one?
Not to start, but someone should read what it writes before it handles logins, payments or other people’s data. That is this site’s advice; the reasons are on what vibe coding means and the build an app with AI cheat sheet.
Where to go next
- “What is an agent, exactly?” Read what is an AI agent? and AI agents for beginners.
- “How do I brief one?” See prompting agents.
- “What should it be allowed to do?” See agent permissions and prompt injection.
- “What else goes into an app?” See the AI coding stack and the security checklist.
- “How do I add skills or connect tools?” See how to install AI skills and what is MCP?
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What happened to Gemini CLI for most individual users in June 2026?
- Google made it free for everyone
- Google stopped serving it and pointed them to Antigravity CLI
- Google merged it into GitHub Copilot
Show the answer
Answer: Google stopped serving it and pointed them to Antigravity CLI. From 18 June 2026 Google stopped serving Gemini CLI for AI Pro, AI Ultra and free users; Antigravity CLI replaced it, while paid organisations kept Gemini CLI.
2. What is the main difference between a coding agent working on your computer and one working in the cloud?
- In the cloud it works on a copy and hands back a change to review
- In the cloud it cannot run any commands
- On your computer it never touches your real files
Show the answer
Answer: In the cloud it works on a copy and hands back a change to review. The page explains that locally the agent works on your real files, while cloud versions work on a copy and return a change you approve.
3. Which of the five tools needs a paid plan or an Anthropic Console account to use?
- Codex
- GitHub Copilot
- Claude Code
Show the answer
Answer: Claude Code. Anthropic says Claude Code is included in all paid Claude plans and marks it No for Free; Codex, Cursor, Copilot and Antigravity CLI have a free way in.
- 1Anthropic, Claude Code overview, read at source 19 Oct 2026“Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser.” ↑anative install for “macOS, Linux, WSL”, “Windows PowerShell” and “Windows CMD”; “Claude Code runs on several surfaces: the terminal, IDE extensions, a desktop app, and the web.” ↑b“Run Claude Code in your browser with no local setup. Kick off long-running tasks and check back when they’re done”; OpenAI, Codex Cloud, read at source 19 Oct 2026: “Each task has its own workspace and can keep working while your computer is asleep.” ↑c“Create skills to package repeatable workflows your team can share, like /review-pr or /deploy-staging.” ↑d“If your repository already has an AGENTS.md for other coding agents, Claude Code can read that in place of a CLAUDE.md.” ↑ethe VS Code tab offers “Install for VS Code” and “Install for Cursor”. ↑f
- 2OpenAI, openai/codex (GitHub README), read at source 19 Oct 2026“Codex CLI is a coding agent from OpenAI that runs locally on your computer.” and “This repository is licensed under the Apache-2.0 License.” ↑
- 3OpenAI, Codex Cloud, read at source 19 Oct 2026“Give Codex coding tasks to work through in the cloud, from investigating bugs to building features.” and “Review changes and continue from the web, mobile, or desktop app.” ↑
- 4Google, Antigravity overview (CLI tab), read at source 19 Oct 2026“A keyboard-first terminal client that runs Antigravity workflows in local shells, remote cloud environments” “and SSH sessions.” ↑
- 5Cursor, Cursor Documentation, read at source 19 Oct 2026“Cursor is a coding agent for building ambitious software.” ↑
- 6Cursor, Agent overview, read at source 19 Oct 2026“Agent is Cursor’s assistant that can complete complex coding tasks independently, run terminal commands, and edit code.” ↑
- 7GitHub Docs, What is GitHub Copilot?, read at source 19 Oct 2026“GitHub Copilot is an AI assistant that helps you write, understand, and ship software.” ↑a“Use Copilot Free to explore core features with no paid plan required.” ↑b
- 8GitHub Docs, About GitHub Copilot cloud agent, read at source 19 Oct 2026“Copilot cloud agent can research a repository, plan changes, and implement them in the background.” ↑a“It can edit files and run tests and linters in an ephemeral cloud development environment, rather than your local IDE.” “You can start work from the agents panel, a conversation, or an issue or pull request on GitHub.com.” ↑b
- 9Google, Transitioning Gemini CLI to Antigravity CLI, 19 May 2026, read at source 19 Oct 2026“On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra”, also for those “using it free of charge using Gemini Code Assist for individuals.” ↑a“your access remains unchanged”; “Gemini CLI will remain accessible via paid Gemini and Gemini Enterprise Agent Platform API keys.” ↑bAntigravity CLI keeps “Agent Skills, Hooks, Subagents, and Extensions” (now “Antigravity plugins”); “there won’t be 1:1 feature parity right out of the gate.” ↑c
- 10Google, Gemini CLI migration, read at source 19 Oct 2026“Antigravity CLI preserves backward compatibility with the core developer-experience constructs popularized by Gemini CLI.” The agent “continues to parse and enforce rule constraints” in workspace rule files (GEMINI.md and AGENTS.md). ↑
- 11OpenAI, Codex IDE extension, read at source 19 Oct 2026“VS Code and compatible editors use the Codex extension; Xcode and JetBrains IDEs provide their own integrations.” ↑
- 12OpenAI, Quickstart (ChatGPT learn docs), read at source 19 Oct 2026Codex is chosen in the ChatGPT desktop app for “software development with codebase context and developer tools”; “The ChatGPT desktop app is available for macOS, Windows, and Linux.” ↑
- 13Cursor, Install, read at source 19 Oct 2026install steps for Mac, Windows and Linux; Cursor, CLI overview, read at source 19 Oct 2026: “Cursor CLI lets you interact with AI agents directly from your terminal to write, review, and modify code.” ↑
- 14GitHub Docs, Copilot feature matrix, read at source 19 Oct 2026agent mode ticked for VS Code, Visual Studio, JetBrains, Eclipse and Xcode; “The GitHub Copilot feature matrix is currently in public preview and is subject to change.” GitHub Docs, About GitHub Copilot CLI, read at source 19 Oct 2026: “Linux”, “macOS”, “Windows from within Powershell and Windows Subsystem for Linux (WSL)”. ↑
- 15Cursor, Cloud Agent, read at source 19 Oct 2026“Cloud agents use the same agent fundamentals but run in isolated VMs in the cloud with full development environments instead of on your local machine.” Access: “Start and manage agents from the Cursor iOS app”, “from cursor.com/agents on any device”, “Use the @cursor command” in Slack, “Comment @cursor on a GitHub PR or issue”. ↑
- 16Agent Skills, Agent Skills overview, read at source 1 Oct 2026 (as cited on this site’s AI skills explained page): “The Agent Skills format was originally developed by Anthropic, released as an open standard, and has been adopted by a growing number of agent products.”↑
- 17OpenAI, Build skills, read at source 5 Oct 2026 (as cited on this site’s AI skills explained page): “Skills work in Chat and Work across ChatGPT on the web, desktop, and mobile, in Codex in the ChatGPT desktop app, and through Codex CLI.” GitHub Docs, About agent skills, read at source 5 Oct 2026 (same page)skills work with “Copilot cloud agent, Copilot code review, the GitHub Copilot CLI, the GitHub Copilot app, and agent mode in Visual Studio Code and JetBrains IDEs”. Cursor, Agent Skills, read at source 5 Oct 2026 (same page): Cursor “automatically discovers skills from skill directories and makes them available to Agent.” Google, Antigravity overview (CLI tab), read at source 19 Oct 2026: “Custom skills (SKILL.md) & rule files”. ↑
- 18Cursor, Available models, read at source 19 Oct 2026“Cursor offers its own models” “alongside other frontier models from OpenAI, Anthropic, and Google.” ↑
- 19GitHub Docs, Supported AI models in GitHub Copilot, read at source 19 Oct 2026“GitHub Copilot supports multiple AI models, each with different strengths.” “The models available to you depend on your Copilot plan and where you’re using Copilot, such as GitHub.com or an IDE.” Providers in the model table: OpenAI, Anthropic, Google, Microsoft, Moonshot AI, xAI. ↑
- 20Google, Antigravity models, read at source 19 Oct 2026“leading frontier reasoning models from Google DeepMind and partner providers”; the list names Gemini, Claude and GPT-OSS models. ↑
- 21Pearce et al., Asleep at the Keyboard? Assessing the Security of GitHub Copilot’s Code Contributions, arXiv, revised 16 Dec 2021, read at source 1 Oct 2026 (as cited on this site’s vibe coding page): “Of these, we found approximately 40% to be vulnerable.”↑
- 22OpenAI, Pricing (ChatGPT learn docs), read at source 19 Oct 2026“ChatGPT Work and Codex are included in your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan.” ↑
- 23Cursor, Pricing, read at source 19 Oct 2026a plan named Free, with “Limited AI usage”, “Limited agent requests”. ↑
- 24Google, Antigravity plans, read at source 19 Oct 2026“Users not on Google AI Pro or Ultra plans receive the following benefits:” (a weekly allowance and weekly rate limit); all plans include “Access to all product features, such as scheduled tasks and the CLI.” ↑
- 25Anthropic, Claude plans and pricing, read at source 19 Oct 2026“Claude Code is included in all paid plans.” The comparison table marks Claude Code “No” for the Free plan. ↑
Read at source 19 Oct 2026: Anthropic, Claude Code overview and plans; OpenAI, openai/codex, Codex Cloud, Codex IDE extension, Quickstart and Pricing; Google, Transitioning Gemini CLI to Antigravity CLI (19 May 2026), Antigravity overview, Gemini CLI migration, models and plans; Cursor, documentation, Agent, Cloud Agent, CLI, Install, Available models and Pricing; GitHub Docs, What is GitHub Copilot?, cloud agent, Copilot CLI, feature matrix and supported models.
Reused from this site’s AI skills explained page (read 1 and 5 Oct 2026) and vibe coding page (read 1 Oct 2026): the Agent Skills overview, OpenAI’s Build skills, GitHub’s About agent skills, Cursor’s Agent Skills and Pearce et al. (2021).
The builder comparison, the “which one should you pick” list and the advice marked as this site’s reading are this site’s own, not claims from these sources. No tool on this page was tested for this comparison.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK