Quick verdict
Claude Code and OpenAI Codex both cost $20 a month at the tier most people will buy, and both act as coding agents that read your repo, edit files and run commands — the real difference is where the work happens. Claude Code lives in your local terminal and works directly in your environment. Codex leans into the cloud: hand it a task and it works asynchronously in a sandbox while you do something else. Pick Claude Code for long, hands-on, multi-file work in your own setup; pick Codex if you already pay for ChatGPT or want to delegate tasks in parallel.
Bottom line: Same $20 entry price, two different shapes of work. Claude Code sells you a powerful agent inside your terminal, sharing one usage allowance with Claude chat. Codex sells you a delegatable agent across CLI, editor and web, sharing one allowance with ChatGPT. Your existing subscription is the cheapest tiebreaker.
Claude Code vs Codex at a glance
| Feature | Claude Code (Anthropic) | Codex (OpenAI) | Verdict |
|---|---|---|---|
| Cheapest plan that includes it | Claude Pro, $20/mo ($17/mo billed annually) | ChatGPT Plus, $20/mo; limited access on Free and Go ($8/mo) | Codex — a limited free taste exists |
| Power tiers | Claude Max 5x $100/mo, Max 20x $200/mo | ChatGPT Pro from $100/mo, higher tier $200/mo | Tie — identical price ladder |
| Free plan includes the agent? | No | Limited access, rolling out | Codex |
| Where it runs | Your terminal and local environment first, plus editor, desktop and web surfaces | Cloud sandboxes first, plus CLI, IDE extension, web and mobile | Depends on your workflow |
| Working style | Interactive, in your shell, with your files, git history and tools | Asynchronous delegation — assign tasks, review results later | Claude Code for pairing; Codex for delegating |
| Models | Anthropic Claude models only | OpenAI models only | Tie — both are single-vendor |
| Usage limits | Five-hour session windows plus weekly limits, shared with Claude chat | Five-hour windows on Plus (estimates published per model); higher Pro tiers relax this | Codex publishes clearer estimates |
| Pay-as-you-go option | Yes, with an Anthropic API key | Yes, with an OpenAI API key | Tie |
How is Claude Code different from Codex in practice?
Claude Code is Anthropic’s terminal-based coding agent. It runs in your local environment, so it sees your actual files, your git history and your shell, and it edits and runs commands right there. That makes it feel like pairing with a fast colleague who is sitting in your repo: you steer, interrupt and redirect as it works through a multi-file change.
Codex is OpenAI’s agentic coding product, and most of its work happens asynchronously in cloud sandboxes. You describe a task — fix a bug, add a feature, open a pull request — and Codex works on it in an isolated environment while you get on with something else. It also ships as a CLI, an IDE extension and a web app, so local work is covered too, but delegation is the centre of gravity. Its CLI runs inside an operating-system sandbox with network access off by default, which is a sensible default for agent safety.
What does each one actually cost?
Neither tool has its own separate subscription — both ride on their parent chat plans, and in 2026 the ladders are almost identical. Claude Code is included from Claude Pro at $20 a month ($17 a month billed annually); it is not included in Claude’s free plan. Heavy users step up to Claude Max at $100 a month for five times Pro’s usage, or $200 a month for twenty times. Codex is included from ChatGPT Plus at $20 a month, with limited access on the Free plan and the $8-a-month Go plan, and ChatGPT Pro tiers at $100 and $200 a month for much higher usage.
The catch is metering, not headline price. Claude Pro and Max usage is shared between Claude chat and Claude Code, and it is capped in five-hour session windows with weekly limits on top — a long coding session and a heavy chat day draw from the same pot. Codex usage is likewise shared with ChatGPT, with published per-model message estimates for each five-hour window on Plus. Neither vendor promises a fixed token count on individual plans, so treat any exact “hours of coding” figure you see elsewhere with suspicion.
Which one writes better code?
Honestly, that is the wrong first question — and this comparison is research-based, not a hands-on benchmark, so we will not invent one. Both agents run their maker’s frontier models, both are restricted to those models (Claude Code runs only Anthropic models; Codex runs only OpenAI models), and both can read a repo, edit multiple files, run tests and explain their changes. The difference that actually changes your day is workflow fit: local-and-interactive versus cloud-and-delegated. If you are still deciding between agent styles, our Claude Code vs Cursor comparison covers the terminal-agent-versus-AI-editor question, and Cursor vs GitHub Copilot covers the editor-plugin end of the market.
Which should you pick?
Pick Claude Code if you live in the terminal, want the agent working directly in your real environment, and are happy steering it interactively through long, multi-file jobs. It is also the natural pick if you already pay for Claude Pro or Max.
Pick Codex if you already pay for ChatGPT, want a limited free taste before committing, or prefer to fire off tasks to a cloud sandbox and review finished work — including running several tasks in parallel. Its published usage estimates also make capacity planning a little less of a guessing game.
Pick neither yet if a $10 in-editor assistant covers your needs — compare the cheaper editor route first in the Cursor vs Copilot piece linked above.
Frequently asked questions
Is Claude Code or Codex cheaper?
At the tier most people buy, they are the same: $20 a month (Claude Pro, or $17 billed annually, versus ChatGPT Plus). Codex is cheaper to try, with limited access on Free and the $8 Go plan; Claude Code has no free tier.
Can I use Claude Code with a free Claude account?
No. Claude Code starts at Claude Pro ($20 a month). The free Claude plan does not include it, though you can use Claude Code with pay-as-you-go API billing instead of a subscription.
Does Codex work offline in my own repo?
Codex has a local CLI that works in your repo inside a sandbox, but its signature workflow is cloud delegation: tasks run asynchronously in OpenAI’s sandboxes and come back for review. Claude Code is the more local-first of the two.
Can I switch between them easily?
Yes — both work on ordinary git repos and neither locks your code into a proprietary format. The real cost of switching is re-learning the workflow, not migrating your project.