Comparison
Claude Code and Codex CLI are two of the best-known coding agents for the terminal. They do the same job in the same place, so the differences that matter are practical: who makes them, which account you sign in with, how they take instructions and how they ask for permission. Here they are side by side, from each vendor’s own documentation.
Facts about third-party tools last checked 9 October 2026, from each vendor’s own documentation.
Get started| Claude Code | Codex CLI | |
|---|---|---|
| Made by | Anthropic | OpenAI |
| Source code | Proprietary: a public repository, all rights reserved under Anthropic’s Commercial Terms | Open source under the Apache-2.0 licence, written in Rust |
| Models | Anthropic’s Claude models | OpenAI’s models |
| Sign in with | A Claude Pro or Max plan, Claude for Teams or Enterprise, a Claude Console account, or a cloud provider (Amazon Bedrock, Google Cloud, Microsoft Foundry) | A ChatGPT plan (Plus, Pro, Business, Edu or Enterprise), or an API key |
| Project instructions | CLAUDE.md; recent versions read AGENTS.md when a repository has no CLAUDE.md | AGENTS.md |
| Approvals | Permission rules and modes; --dangerously-skip-permissions turns prompts off | /permissions chooses what it may do without asking, within a sandbox you can inspect |
| MCP servers | Supported | Supported (codex mcp) |
| Install | Native installer, Homebrew or WinGet (npm is deprecated) | Install script, npm or Homebrew |
Each one signs in with its own vendor’s plans. If you already have a Claude plan or a ChatGPT plan, that usually decides which agent costs you nothing extra to try.
Codex CLI’s source is open, so you can read it, build it and change it. Claude Code’s repository is public, but its code is not under an open-source licence.
Codex reads AGENTS.md. Claude Code reads CLAUDE.md, and recent versions read AGENTS.md instead when a repository has no CLAUDE.md, so one AGENTS.md can serve both. If you also keep a CLAUDE.md, add the line @AGENTS.md to it: Claude Code reads only the CLAUDE.md then, and the line imports the shared file.
They run different model families, which tend to get different things wrong. That is a reason to run both on the same work, not just to pick one.
You don’t have to choose. In a YOLO Studio workspace both come pre-installed: Claude Code with your Claude plan or an API key, Codex with your ChatGPT plan or an API key. Start each in its own lane, a separate git clone on its own branch, and they work on the same repository without colliding.
Then put the difference in model families to work: let one agent build and the other review. The Background Runner’s Agent review requires the reviewer to run a different model from the one that did the work.
It depends on the task, the codebase and which models are current, and that changes faster than any comparison page. Try both on your own repository; in YOLO Studio you can run them side by side on the same task.
Not safely in one working copy, because they would edit the same files. Give each its own checkout: a git worktree locally, or a lane in YOLO Studio.
Sign in with GitHub, open a workspace on your repository, and put your first agent to work in a few minutes.
Get started
