What is the difference between Copilot and Codex?
GitHub Copilot is an AI assistant spread across your IDE and GitHub; OpenAI Codex is a coding agent you hand tasks to. Copilot started as autocomplete and now adds agent mode in the editor, a terminal CLI, a cloud agent that works from GitHub issues and code review on pull requests, with models from Anthropic, OpenAI, Google and others. Codex runs in a terminal, an IDE extension, the ChatGPT desktop app and the cloud, on OpenAI's models, and is included in ChatGPT plans.
The two also overlap more than the names suggest. Copilot can run Codex as a third-party agent on GitHub, and it offers OpenAI's Codex models in its model picker. So the choice is partly about which subscription you already have, and partly about whether you want suggestions as you type or an agent you delegate to.
How does GitHub Copilot work?
GitHub Copilot is "an AI assistant that helps you write, understand, and ship software." Its main surfaces:
- In your IDE: code completion and next edit suggestions as you type, plus agent mode, in VS Code, Visual Studio, JetBrains, Eclipse and Xcode.
- Copilot cloud agent: works "in an ephemeral cloud development environment, rather than your local IDE," and opens a pull request for each task. You start it from the agents panel, a conversation, an issue or a pull request.
- Copilot CLI: the
copilotcommand, with plan mode,-pfor one-shot tasks,/modelto switch models and/sandbox enablefor local sandboxing. - Copilot code review on pull requests, which can follow your instruction files.
- Third-party coding agents, including OpenAI Codex, in public preview.
How does Codex work?
Codex CLI lets you "inspect code, make changes, run commands, and automate repeatable work without leaving your terminal." The openai/codex repo is open source under Apache-2.0. You run codex in a project and sign in with ChatGPT or another method, such as an API key. The same agent runs in several places:
- Codex CLI, with
/permissionsto choose what it may do without asking andcodex execfor scripts and pipelines. - Codex IDE extension for VS Code, Cursor, Windsurf and VS Code Insiders. Xcode and JetBrains IDEs "provide their own integrations," per the IDE page.
- The ChatGPT desktop app for macOS, Windows and Linux, where Codex is one of the two options when you start.
- Codex Cloud, which runs tasks in cloud environments you set up, started from the web, the desktop app, mobile or the CLI. Tasks "can keep working while your computer is asleep."
- GitHub code review, where a comment with
@codex reviewon a pull request gets a standard GitHub review.
OpenAI's models page recommends GPT-6.1 Sol for complex coding and agentic work, and says to "Use Luna for focused, repeatable tasks." The Codex CLI guide covers setup.
Copilot vs Codex side by side
| GitHub Copilot | OpenAI Codex | |
|---|---|---|
| Made by | GitHub | OpenAI |
| Main way of working | Suggestions as you type, plus agents | Tasks you delegate to an agent |
| IDE agent | Agent mode in VS Code, Visual Studio, JetBrains, Eclipse, Xcode | Extension for VS Code, Cursor, Windsurf; own integrations in Xcode and JetBrains |
| Terminal agent | copilot |
codex (open source, Apache-2.0) |
| One-shot scripting | copilot -p |
codex exec |
| Cloud agent | Copilot cloud agent on GitHub | Codex Cloud |
| PR review | Copilot code review | @codex review and automatic reviews |
| Models | Claude, GPT, Gemini, Grok and others, by plan | OpenAI models such as GPT-6.1 Sol and GPT-6 Luna |
| Project instructions | .github/copilot-instructions.md, .github/instructions/, plus AGENTS.md, CLAUDE.md or GEMINI.md on some surfaces |
AGENTS.md (and AGENTS.override.md) |
| Project skills folders | .github/skills/, .claude/skills/, .agents/skills/ |
.agents/skills/ |
| Comes with | A Copilot plan | A ChatGPT plan, or an API key without cloud features |
| Entry paid price | Pro, $10 a month | ChatGPT Go, $8 a month |
How do the cloud agents and reviews compare?
Cloud agents. Copilot cloud agent lives inside GitHub. You assign an issue to Copilot, and it handles the branch, commits and pull request. Codex Cloud starts from an environment: you pick repositories, Codex installs dependencies and tests the setup, and you publish it. Each task then gets its own workspace, and when it finishes you review the changes and open a pull request.
Code review. Both review pull requests on GitHub. Copilot code review reads .github/copilot-instructions.md, path instructions, AGENTS.md and a REVIEW.md file. Codex's GitHub integration reports only "P0 and P1 issues," and you can steer it with a "## Code Review Rules" section in the AGENTS.md closest to the code.
Codex inside Copilot. GitHub's third-party coding agents page says the integration uses "the Codex SDK and can be powered by your existing Copilot subscription." Turning on "Allow Codex coding agent" installs a GitHub App, and you can then assign issues to Codex or mention it in pull request comments. Sessions use GitHub Actions minutes and AI credits. The Codex VS Code extension also has a "Sign in with Copilot" option for Pro+ and Max subscribers.
How do instructions and skills carry over?
AGENTS.md. Codex reads AGENTS.md from ~/.codex for global guidance, then from each folder between the project root and your current directory, with closer files taking priority. It stops adding files at 32 KiB by default, set by project_doc_max_bytes. Copilot reads AGENTS.md on cloud agent, Copilot CLI, code review on GitHub.com and VS Code Chat, alongside its own .github/copilot-instructions.md. A shared AGENTS.md at the repo root works in both.
Skills. Codex loads skills from .agents/skills in the repo and $HOME/.agents/skills, and you can run /skills or type $ to mention one. Copilot's agent skills page lists .github/skills, .claude/skills or .agents/skills for projects, and ~/.copilot/skills or ~/.agents/skills for personal skills. So .agents/skills/ is the folder both tools share. The Codex skills guide covers the format.
How much do Copilot and Codex cost?
Prices below are as of October 2026.
- GitHub Copilot (plans): Free, limited to 2,000 completions a month; Pro at $10 a month with 1,500 AI credits; Pro+ at $39 with 7,000; Max at $100 with 20,000; Business at $19 per seat and Enterprise at $39 per seat. Completions are unlimited on paid plans, and AI credits are spent by tokens.
- Codex (pricing): included in ChatGPT Free and Go ($8 a month) for lighter coding tasks; Plus at $20 a month with Codex on the web, CLI, IDE extension and iOS; Pro from $100 a month, with $100, $200 and $500 options; Business at $20 per user per month billed annually or $25 billed monthly; Enterprise and Edu through sales. With an API key you pay API rates and get Codex in the CLI, SDK or IDE extension, without cloud features.
Which should you pick?
GitHub Copilot tends to fit when:
- You want suggestions as you type, plus an agent when you need one.
- Your work starts in GitHub issues and you want them turned into pull requests.
- You want to switch between Claude, GPT, Gemini and other models in one subscription.
- Your company already pays for Copilot Business or Enterprise.
Codex tends to fit when:
- You prefer handing whole tasks to an agent and reviewing the diff.
- You already pay for ChatGPT and want coding included.
- You want an open source CLI you can inspect, script with
codex execor run with an API key. - You want cloud tasks that keep running from your phone or the desktop app.
You can also combine them: keep Copilot for completions and run Codex as a third-party agent on GitHub, with one AGENTS.md and one .agents/skills/ folder serving both.
Keep your skills in one place
The skills you refine are worth more than any one subscription. Kipt keeps your prompts and skills as Markdown in a private GitHub repo you own, and installs every skill into Claude Code or Codex with npx skills add https://github.com/<you>/<repo>.git --all -g. To turn a prompt into a skill folder, try the free SKILL.md generator.