Short answer
If you already pay for a ChatGPT plan and do not want another subscription, use Codex. If you have a Claude subscription, or you prefer how Claude models answer, use Claude Code. By the official docs, the two tools show up in almost the same places (terminal, editor extension, desktop, web or cloud), and both have rules files, MCP, skills, hooks and subagents. A feature list will not settle it. What settles it is which company's account you already pay for, and which model suits your work.
Side by side
That almost every cell is filled for both is the finding. The differences show up in the situations below.
Where you use it
- Codex (OpenAI)
- ChatGPT desktop, mobile and web; Codex CLI; IDE extension; Codex cloud
- Claude Code (Anthropic)
- Terminal, VS Code and JetBrains extensions, desktop app, web and mobile app
Sign-in
- Codex (OpenAI)
- ChatGPT account
- Claude Code (Anthropic)
- Claude subscription or Anthropic Console account
Rules file
- Codex (OpenAI)
- AGENTS.md (a global file and per-folder files, joined from the root down)
- Claude Code (Anthropic)
- CLAUDE.md. It can read AGENTS.md instead if that exists
Permissions
- Codex (OpenAI)
- Ask for approval, auto-review approval, or allow everything
- Claude Code (Anthropic)
- Manual mode is the default. It asks before file edits and shell commands. In auto mode a classifier reviews them
Cloud runs
- Codex (OpenAI)
- Pick a GitHub repository and work in a cloud environment. Keeps going if your computer sleeps. You review, then open a PR
- Claude Code (Anthropic)
- Runs on the web. Scheduled tasks run even when your computer is off
Extending
- Codex (OpenAI)
- Skills, MCP, hooks, subagents, GitHub Actions
- Claude Code (Anthropic)
- Skills, MCP, hooks, subagents, Agent SDK, GitHub Actions
Pricing
- Codex (OpenAI)
- Included in ChatGPT plans (see the official pricing page)
- Claude Code (Anthropic)
- Included in Claude plans (see the official pricing page)
When your subscription decides for you
If you use ChatGPT Plus and turn on Codex, you start in the same app with no separate signup. Codex's docs say the latest model is available in Codex on the Plus, Pro, Business, Enterprise and Edu plans. I could not confirm in the docs whether the free plan is included.
If you already have Claude Pro or Max, Claude Code comes with that subscription. Usage limits differ by plan, and the people who hit them first are the ones who hand over long jobs. This article does not give limit numbers for either service. The docs are updated on different schedules, so any figure would likely be wrong. Check each official pricing page.
Scenario 1: a job you throw over the wall at the end of the day
Both can run in the cloud. In Codex cloud you pick a repository, and it sets up the environment, installs dependencies and runs tests. Each task gets its own workspace, so several can run at once. When one finishes, you review the changes and test results and then open a PR.
Claude Code's docs describe starting a long job on the web and pulling it into the terminal later. Cloud scheduled tasks run even with your computer off, and can start from GitHub events. If you have a repeating job like reviewing PRs every morning, its docs are more specific about that.
Scenario 2: a folder with files you cannot lose
Codex recommends starting with its ask-for-approval mode. It works freely inside the current folder and stops to ask when it would cross the boundary. The auto-review mode hands those same boundary checks to an automatic reviewer instead of asking you.
In Claude Code's manual mode, reading inside the working folder needs no prompt, but file edits and most shell commands ask every time. You can have it remember a command you allowed, per repository and per command. Both feel slow and annoying at first. Skipping this stage and starting with allow-everything is how accidents happen.
Scenario 3: a team using both in one repository
The rules file names differ. Codex uses AGENTS.md and Claude Code uses CLAUDE.md. If you keep only AGENTS.md in the team repository, Claude Code can read it, so keeping rules in one place is easier to maintain. Codex's official docs do not say it reads CLAUDE.md.
Codex reads AGENTS.md up to a combined 32 KiB. A team that writes long rules will see the tail end cut off at that limit.
What trips people up when switching
- Chat history and settings do not carry over. Only the rules file and your list of MCP connections can be moved, by hand.
- Slash command names are different, so commands you memorized will not work as they are.
- The same request gets answers with different habits. Do not accept a general claim that one model is better until you have tried it on your own work.
Give both the same task
Before you pay, use a free trial or a short stretch of use, put the same request into a copy of the same folder, and compare.
Read this repository and add a "How to run it for the first time" section to the README, five lines or fewer. Actually run the commands and write down only what works. If a step does not work, do not put it in the README. Tell me about it separately.
Watch for: whether it really ran the commands, and whether it reports a failing step instead of making one up. This request separates writing something plausible from checking before writing.
FAQ
Can I use Codex outside a terminal?
Yes. The official docs list ChatGPT on desktop, mobile and web, the Codex CLI, an IDE extension and Codex cloud as places to use it.
Are the rules files compatible?
Claude Code's docs say it can read AGENTS.md in place of CLAUDE.md. Codex uses AGENTS.md, and I could not find any statement that it reads CLAUDE.md.
Which one is safer?
Both have approval modes, and safety depends on your settings. Start in a mode that asks, and only hand over a folder you cannot afford to lose when you have a backup.
I cannot code. Which is easier?
The places you can use them overlap so much that it is hard to call one easier. Starting from the chatbot app you already use (Codex if ChatGPT, Claude Code if Claude) cuts the installation steps.
Keep going
Checked against sources on 2026-11-05. Sources: Codex docs (home), Codex: permission modes, Codex: cloud, Codex: AGENTS.md, Claude Code docs: Overview, Claude Code docs: Permissions. How we write