Claude Code vs Codex: which coding agent?
Both work in your terminal on your real files, and both come with a plan you may already pay for. I run both. Where each one wins, and how to decide.
By Nick
On this page
Before you start
Claude Code
Choose Claude Code if you pay for Claude, or if you want the agent that is most careful about asking before it acts and best at explaining what it did.
Read my notesCodex
Choose Codex if you pay for ChatGPT, or if you want the same models to also power other tools and image generation from one subscription.
Read my notesBottom line
Start with whichever plan you already pay for. They trade places on features every few months, and the habits that keep you safe (Git, one task at a time, read the diff) are identical. Running both is how I check one against the other.
Side by side
| Claude Code | Codex | |
|---|---|---|
| Included with | Paid Claude plans: Pro, Max, Ultra, Team, Enterprise | Every ChatGPT plan, Free and Go included, with more use on paid plans |
| Cheapest plan that includes it (6 Nov 2026) | Claude Pro, $20 a month ($17 a month paid yearly) | ChatGPT Free for a trial; Go $8 or Plus $20 a month for real use |
| Bigger plans (6 Nov 2026) | Max $100 (5 times Pro), Ultra $200 (10 times Pro each week) | Pro at $100, $200 or $500 a month |
| Where it runs | Terminal, editor extension, desktop app, phone | Terminal, browser sandbox, editor extension |
| Plan-first mode | Built in; read the plan, then say go | Ask for it in the prompt |
| Permissions | Permission modes, switched per session with Shift+Tab | Approval levels, per session |
| Standing instructions | CLAUDE.md | AGENTS.md |
| Skills and routines | Skills, hooks, scheduled routines | Growing; check current docs |
| What else the plan buys | The Claude app | The ChatGPT app, image generation, models other tools can call |
| Usage limit shape | A five-hour limit plus a weekly cap, shared with the Claude app | A five-hour limit plus weekly limits on Plus; Pro has no five-hour limit |
| When you run out | Wait, upgrade, or pay for usage credits at API rates | Wait, switch to a cheaper model, upgrade, or buy credits |
| Where to check what is left | Settings, then Usage, or /usage in Claude Code | The Codex usage dashboard, or /status in the Codex CLI |
- Included with
- Cheapest plan that includes it (6 Nov 2026)
- Bigger plans (6 Nov 2026)
- Where it runs
- Plan-first mode
- Permissions
- Standing instructions
- Skills and routines
- What else the plan buys
- Usage limit shape
- When you run out
- Where to check what is left
Both of these do the same job: you describe an outcome, they read your project, change the files, run the thing and come back. Both are agents, not chat windows. Both are included with a plan you may already have.
I use both every day. The honest comparison is short.
The question that decides it
Which plan do you already pay for? Claude plan, use Claude Code. ChatGPT plan, use Codex. That settles it for most people, and it is the right answer, because the thing that makes a coding agent go well is not the agent. It is Git, one task at a time and reading what it did, and those are the same in both.
Where Claude Code wins, for me
It asks before it acts, visibly, with a plan mode and permission modes I actually use. Its explanations of what it changed are the clearest. Skills, hooks and scheduled routines have made it the one I build on.
Where Codex wins, for me
One subscription does more. The ChatGPT plan that includes Codex also draws my still images, and the models behind Codex are what my scheduled agent calls when a job needs thinking. Codex comes with ChatGPT is the longer version of why that matters to the bill. The browser sandbox is also the gentlest first step if a terminal (a window where you type commands) puts you off.
What I actually do
Claude Code for anything that touches this site or needs care. Codex as the second opinion and the worker my agent hands jobs to. When one is confidently wrong, the other usually notices. My setup shows the wiring.
If you have neither plan, pick on the chat app you would rather use for everything else, because that is what you will use most. If you have both, run both for a week on the same small task and keep the one you found yourself trusting.
Common questions
Which one is better at code?
Should I run both?
Which is cheaper, Claude Code or Codex?
What about Cursor?
What to do next
Related reading
Setups
My current AI setup