Codex vs Claude Code: the differences, and how to use both

The short answer

Claude Code is Anthropic's coding agent. Codex is OpenAI's. A coding agent is an AI tool that reads your code, changes files and runs commands in your project. Both do that. They differ in three places:

• The plan you need. Claude Code needs a paid Claude plan or an Anthropic Console account. Codex comes with ChatGPT plans, even the Free plan (as of 26 October 2026).
• The instructions file it reads. Claude Code reads CLAUDE.md. Codex reads AGENTS.md.
• Where it runs. Both have a terminal tool, editor add-ons, a desktop app and a web version. They package them in different ways.

Already pay for one of the two? Start there. Want both? They can share one instructions file. See below for how.

Side by side

Plans and prices change often. This table shows the official pages as they were on 26 October 2026.

Claude Code in brief

Anthropic's overview calls Claude Code an agentic coding tool. It reads your code, edits files and runs commands. It also works with your other developer tools. You can use it in the terminal, in VS Code and JetBrains, in a desktop app and in the browser.

You need a Pro, Max, Team, Enterprise or Console account. The free Claude plan does not include Claude Code, says the setup page. On Windows it runs on Windows 10 1809 or later. It can also run inside WSL, which is Linux running inside Windows.

Your project rules go in a file called CLAUDE.md. Claude reads it at the start of every session, as the CLAUDE.md docs explain.

Codex in brief

OpenAI's Codex docs list where you can use Codex:

• the ChatGPT desktop and mobile apps
• ChatGPT on the web
• the Codex CLI, a tool you run by typing commands in a terminal
• the Codex extension for code editors
• Codex Cloud

OpenAI's pricing page says Codex comes with these ChatGPT plans: Free, Go, Plus, Pro, Business, Edu and Enterprise. You can also use it with an API key, which you pay for per use at API rates. With an API key you don't get the cloud features, such as GitHub code review.

The Codex CLI is open source, under the Apache-2.0 licence. Its README on GitHub calls it a coding agent "that runs locally on your computer". On Windows, the Codex desktop app runs its agent in PowerShell by default. You can switch it to WSL2.

Codex reads its instructions from AGENTS.md files. First it reads a global one in your Codex home folder. Then it reads one per folder, from the top of the project down to where you work. Together they may be 32 KiB at most by default, says the AGENTS.md page in the Codex docs.

How to choose

• You already pay for ChatGPT: try Codex first. It's in your plan.
• You already pay for Claude Pro or Max: Claude Code is included.
• You want to read the tool's source code: the Codex CLI is open source, under Apache-2.0.
• You work on Windows: both work. Claude Code installs from PowerShell or CMD. Codex has a Windows app from the Microsoft Store. The Windows guide covers both.

Neither choice locks you in. Your project is still a folder of files. Either agent can open it tomorrow.

How to use both on one project

Maybe you use one for work and one for your own projects. Or you use one as a second opinion. Two things keep that easy.

1. One set of instructions

Write the rules both agents share in AGENTS.md. Codex reads it. Claude Code reads that same AGENTS.md when the project has no CLAUDE.md. Want some rules for Claude only? Then keep a short CLAUDE.md. Its first line is @AGENTS.md, which pulls in the shared file. Below that line, add the Claude-only rules. The CLAUDE.md docs describe this. Now both agents follow the same rules. The guides to CLAUDE.md and AGENTS.md each have a free template.

2. One record of progress

The two agents don't share a memory. So give them a shared file instead. In my own projects, every run starts from a written prompt file. That file has a "Done when" check, whichever tool runs it. Both Claude Code and Codex then update the same project/status.json in that folder: what's done, what's next, what's waiting on me. My overview doesn't care which tool did the work.

Don't let them edit the same files at once

Two agents that write the same file at the same time will undo each other's work. Run them one after the other. Or give each one its own copy of the project. Anthropic's best practices suggest git worktrees for this: separate copies of your project that git keeps track of. Then each session's edits stay apart.

Common questions

Is Codex free?

OpenAI's pricing page says Codex comes with ChatGPT Free and Go, and with the paid plans, as of 26 October 2026. How much you can use it depends on your plan. Check the pricing page for yours.

Does Claude Code work with the free Claude plan?

No. Anthropic's setup page says the free claude.ai plan does not include Claude Code.

Can Claude Code read AGENTS.md?

Yes, in recent versions. It reads AGENTS.md when the project has no CLAUDE.md. A CLAUDE.md can also pull it in with the line @AGENTS.md.

My take: the bigger difference is the chat, not the code

Both tools write good code. The bigger difference I notice is how ChatGPT and Claude talk with you about your project.

ChatGPT is often very excited about your ideas. That feels good. But it can also send you far down the wrong road before you notice.

Claude is more down to earth. It asks more questions. It pushes back before you make a big call.

So for strategy and action plans, I prefer Claude. Can ChatGPT work too? Yes. If you know your field well, you will spot a bad direction fast. But if a lot of the work is new to you, Claude is the safer pick for planning.