ComparisonUpdated

Codex vs Claude Code in 2026

OpenAI's and Anthropic's coding agents now do much the same work: they read a codebase, change files, run commands and open pull requests, from a terminal, an editor, a desktop app or the browser. What separates them is the plans they come with, how their limits work, and the models they run.

The short answer: if you already pay for ChatGPT, or want to start free, pick Codex. It comes with ChatGPT's plans, and its CLI, editor extension and cloud tasks are listed from Plus, at $20 a month. If you want Claude's models, or already pay for Claude, pick Claude Code, from Claude Pro at $20. Both read AGENTS.md, and Codex can import a Claude Code setup.

By the Fabricate teamWe make Fabricate9 min readPrices checked

The short answer

  • Compared

    Codex

    Best for: a free start, cloud tasks and GPT models

    Codex is OpenAI's coding agent.

  • Compared

    Claude Code

    Best for: Claude's models, on the Claude plan you have

    Claude Code is Anthropic's coding agent.

  • We make this

    Fabricate

    Best for: an app built for you, with no code to write

    Codex and Claude Code make people who write code faster.

Key takeaways

  • Codex and Claude Code do much the same work: they read a codebase, change files, run commands and tests, and work with GitHub, from a terminal, an editor, a desktop app or the browser.
  • Both cost $20 a month to start in full: ChatGPT Plus for Codex, Claude Pro for Claude Code. Codex also runs in ChatGPT's desktop app on the Free and Go plans, as it rolls out.
  • Both count use in five-hour and weekly limits; Codex's Pro currently has no five-hour limit, and Claude's Max gives 5x or 20x Pro's usage.
  • Codex runs OpenAI's GPT models and Claude Code Anthropic's Claude models; through the API, GPT-6.1 Sol and Sonnet 5.5 cost the same.
  • Both read AGENTS.md, and the Codex CLI imports a Claude Code setup, so you can try both on one project.

Compare at a glance

Feature by feature, with the tool that does each one better.

  • Where you use itBoth come to the editor you have; Codex reaches Xcode and JetBrains through their own integrations
    Codex
    ChatGPT's desktop app, a CLI, VS Code, Cursor and Windsurf, the web, iOS
    Claude Code
    A terminal, VS Code, JetBrains, a desktop app, the web, the Claude app
    Winner
    Tie
  • Cheapest way inCodex's CLI, editor extension and cloud tasks start on Plus, at the same $20 as Claude Pro
    Codex
    Free and Go ($8), in the desktop app as it rolls out
    Claude Code
    Claude Pro, $20 a month; Free has no Claude Code
    Winner
    Codex
  • For a full day of codingCodex Pro currently has no five-hour limit; Max gives 5x or 20x Pro's usage
    Codex
    Pro: $100, $200 or $500 a month
    Claude Code
    Max: $100 or $200 a month
    Winner
    Tie
  • Usage limitsNeither publishes a fixed number of messages
    Codex
    Five-hour windows on Plus and Business; weekly limits may apply
    Claude Code
    A five-hour window and weekly limits, shared with Claude chat
    Winner
    Tie
  • Past the limitsBoth let you pay by use rather than wait for the reset
    Codex
    ChatGPT credits on Plus and Pro, or an API key
    Claude Code
    Usage credits at API rates, or a bigger plan
    Winner
    Tie
  • ModelsEach runs its own company's models; with an API key, Codex runs those your key can use
    Codex
    OpenAI's, such as GPT-6.1 Sol and GPT-6 Luna
    Claude Code
    Anthropic's Claude models
    Winner
    Tie
  • Cloud tasksBoth keep working while your computer is off
    Codex
    From your GitHub repositories, on Plus and up
    Claude Code
    On the web, and scheduled tasks in the cloud
    Winner
    Tie
  • Pull requests and CIBoth review pull requests and run unattended in CI
    Codex
    GitHub code review and @codex tasks; codex exec in scripts and CI
    Claude Code
    Opens pull requests; GitHub Actions, GitLab CI/CD and Code Review
    Winner
    Tie
  • Project instructionsBoth read AGENTS.md, so one file can guide both
    Codex
    AGENTS.md
    Claude Code
    CLAUDE.md, or AGENTS.md
    Winner
    Tie
  • Hosting what you buildSites hosts websites and web apps made in ChatGPT or Codex
    Codex
    Sites, on eligible plans, in public beta
    Claude Code
    No: you deploy it where you choose
    Winner
    Codex

When to pick which

  • Codex

    Codex is OpenAI's coding agent. It reads a repository, edits files, runs commands and tests, and reviews code, in the ChatGPT desktop app, a terminal (the Codex CLI), an extension for VS Code, Cursor and Windsurf, on the web and on iOS, and in Xcode and JetBrains through their own integrations. Cloud tasks run from your GitHub repositories, each in its own workspace, and keep going while your computer sleeps, then hand back changes to commit or a pull request. It reads AGENTS.md, connects to other tools through MCP, runs subagents and scheduled tasks, and comes with ChatGPT's plans, on OpenAI's GPT models. The CLI and its SDK are open source.

    Choose it if

    • You already pay for ChatGPT, or want to try a coding agent on its free plan first
    • You want OpenAI's GPT models, such as GPT-6.1 Sol
    • You want a CLI you can read and build on: the Codex CLI and SDK are open source
    • You want to host what you build with it, through Sites on an eligible ChatGPT plan
    • You code all day and want no five-hour limit, which Pro offers for now
  • Claude Code

    Claude Code is Anthropic's coding agent. It reads your codebase, edits files, runs commands and works with your other tools, and it runs in the terminal, in VS Code and JetBrains, in a desktop app and in the browser, including on a phone. It works with git itself: it stages changes, writes commit messages, creates branches and opens pull requests. It reads a CLAUDE.md file (or an AGENTS.md) at the start of every session, connects to other tools through MCP, and can split a job across several agents. It comes with Claude's paid plans and runs Claude's models.

    Choose it if

    • Claude is the model you want, without paying API prices on top of a subscription
    • You already pay for Claude Pro or Max, which include Claude Code
    • Your team is on Claude Team, where every seat includes Claude Code
    • You build on the Claude API too: Max comes with $100 or $200 a month of API credits

How Codex and Claude Code differ now

The two now overlap almost entirely. Both are agents you hand a task to: they read the project, plan, change files, run commands and tests, and hand back changes to review or a pull request. Both run in a terminal, in VS Code, in a desktop app, in the browser and on a phone, both keep working in the cloud while your computer is off, and both read AGENTS.md, connect other tools over MCP and split work across subagents.

The differences start with the plans. Codex comes with ChatGPT's, so it reaches further down: Free and Go, at $8, include it in the desktop app on GPT-6 Luna as it rolls out, while the CLI, the editor extension and cloud tasks are listed from Plus at $20. Claude Code starts at Claude Pro, also $20, or $17 billed yearly. At the top, ChatGPT Pro costs $100, $200 or $500 a month with no five-hour limit for now, and Claude Max $100 or $200, with 5x or 20x Pro's usage and monthly API credits for your own apps.

Then each has its own edges. Codex's CLI and SDK are open source, Sites can host what you build with it on eligible plans, and its CLI imports a Claude Code setup in one command. Claude Code comes with every paid Claude plan, every Team seat included, has its own JetBrains plugin and runs Claude's models. If you can, try both on the same task: one AGENTS.md file guides both.

Which for which job

  • You already pay for one of them

    Start with the agent your plan includes: Codex with ChatGPT Plus, Pro or Business, Claude Code with Claude Pro, Max or Team. Each costs nothing more until you reach its limits.

    Either works

  • Trying a coding agent without paying

    ChatGPT's Free plan includes Codex in the desktop app on GPT-6 Luna, as it rolls out. Claude's free plan has no Claude Code.

    Pick Codex

  • A review on every pull request

    Both review pull requests on GitHub: Codex on Plus and up, and Claude Code through its GitHub Code Review and GitHub Actions, which it also runs in GitLab CI/CD. Codex runs in other pipelines through codex exec.

    Either works

  • Long coding days, every day

    Codex's Pro currently has no five-hour limit, from $100 a month. Claude Code's Max gives 5x or 20x Pro's usage for $100 or $200, within weekly limits.

    Pick Codex

  • Trying both on the same project

    Both read AGENTS.md, so one file of project instructions guides both. The Codex CLI's /import brings a Claude Code setup across, with its skills, MCP servers, hooks, subagents and up to 50 recent chats, and leaves the original as it was.

    Either works

What they cost

  • Codex

    Free
    Codex in the ChatGPT desktop app for quick coding tasks, on GPT-6 Luna, as it rolls out.
    Go, $8 a month
    The same, for lightweight coding tasks.
    Plus, $20 a month
    The CLI, your editor, the web and iOS, cloud tasks and GitHub code review, on GPT-6.1 Sol and GPT-6 Luna. ChatGPT credits extend it past the limits.
    Pro, $100 a month
    Also $200 or $500 a month, with no five-hour limit for now, and Ultrafast mode on Pro $500.
    Business, $25 a user a month
    $20 a user billed yearly, from 2 users, with single sign-on and larger cloud machines.
  • Claude Code

    Free
    No Claude Code: it needs a paid Claude plan.
    Pro, $20 a month
    $17 a month billed yearly ($200 a year). Claude Code included, sharing usage limits with chat.
    Max 5x, $100 a month
    5x Pro's usage per five-hour session, and $100 a month of API credits.
    Max 20x, $200 a month
    20x Pro's usage per five-hour session, and $200 a month of API credits.
    Team and Enterprise
    Team seats, and Enterprise at $20 a seat a month plus usage at API rates, billed yearly.

At $20 a month they cost the same: ChatGPT Plus brings Codex's CLI, editor extension and cloud tasks, and Claude Pro brings Claude Code, or costs $17 a month billed yearly. Below that, only Codex has a way in, in ChatGPT's desktop app on Free and on Go at $8, as it rolls out. Above it, Claude's Max costs $100 or $200 a month and ChatGPT's Pro $100, $200 or $500, with no five-hour limit for now. Neither prices by the request: both count use against five-hour and weekly limits, and both let you keep going past them, Codex with ChatGPT credits and Claude Code with usage credits at API rates. Through the API, GPT-6.1 Sol and Claude Sonnet 5.5 both cost $2 per million tokens in and $10 out.

Where Fabricate fits

We make Fabricate

If what you want is the app, not the code

Codex and Claude Code make people who write code faster. If you would rather not write any, Fabricate builds a web app from a sentence: the screens, a database, sign-in and the server code, in one project you change by asking. Building and previewing are free; publishing is on Pro and Max.

  • Describe the app in plain words: no editor, terminal or setup
  • Every app is opened in a real browser and checked before it can be published
  • On Pro and Max, see, edit and download the code, or send it to GitHub, and carry on in Codex or Claude Code
Try Fabricate free

Build free, go live on Pro.

Frequently asked questions

Is Codex better than Claude Code?
Neither wins outright, since they now do much the same work: both read a codebase, change files, run commands, work with GitHub and run tasks in the cloud. Codex has a way in on ChatGPT's free plan, an open-source CLI, and Pro with no five-hour limit for now. Claude Code runs Claude's models and comes with every paid Claude plan. If you already pay for ChatGPT or Claude, start with the agent your plan includes.
Which is cheaper, Codex or Claude Code?
At $20 a month they cost the same: ChatGPT Plus for Codex, and Claude Pro for Claude Code, which is $17 a month billed yearly. Codex also comes with ChatGPT's Free plan and Go, at $8, in the desktop app as it rolls out. For heavy use, Claude Max costs $100 or $200 a month and ChatGPT Pro $100, $200 or $500. Through the API, GPT-6.1 Sol and Claude Sonnet 5.5 both cost $2 per million tokens in and $10 out.
Is Codex free?
Partly. ChatGPT's Free plan includes Codex in the desktop app for quick coding tasks on GPT-6 Luna, as it rolls out, and Go, at $8 a month, for lightweight ones. The CLI, the editor extension, Codex on the web and cloud tasks are listed from Plus, at $20. Claude Code has no free plan: it starts with Claude Pro.
Can I use Codex and Claude Code together?
Yes. Both read AGENTS.md, so one file of project instructions can guide both, and both run in VS Code and in a terminal. The Codex CLI can also import a Claude Code setup with /import: instruction files become AGENTS.md, and settings, skills, MCP servers, hooks and subagents come across, with up to 50 chats from the last 30 days. The Claude Code setup is left as it was.
What are the usage limits on Codex and Claude Code?
Neither gives a fixed number of messages. On ChatGPT Plus, OpenAI estimates about 15 to 160 local Codex messages every five hours on GPT-6.1 Sol, with weekly limits that may apply, and Pro currently has no five-hour limit. Claude Code shares Claude's limits with chat: a rolling five-hour window plus weekly limits, with Max giving 5x or 20x Pro's usage. Past the limits, Codex takes ChatGPT credits and Claude Code usage credits at API rates.
Which models do Codex and Claude Code use?
Each runs its own company's. Codex runs OpenAI's GPT models, with GPT-6.1 Sol recommended for complex work and GPT-6 Luna for focused tasks, and GPT-6 Astra as well. Claude Code runs Anthropic's Claude models. With an API key, Codex runs the models your key can use.
Can Codex or Claude Code host my app?
Codex can, through Sites, which hosts websites and web apps made in ChatGPT or Codex on eligible plans during its public beta, with a database, file storage and Sign in with ChatGPT. Claude Code doesn't host apps: it writes the code, and you deploy it where you choose.
Do I need to know how to code to use either?
Mostly, yes. Both work in a codebase: they read, change and run code, and you review what they did. If you want an app without working in code, an app builder such as Fabricate, which we make, builds it from a description.

Sources

The pages this guide quotes, each with the day it was read.

  1. Codex docs: pricing (opens in a new tab), checked
  2. Codex docs: models (opens in a new tab), checked
  3. Codex docs: Codex CLI (opens in a new tab), checked
  4. Codex docs: IDE extension (opens in a new tab), checked
  5. Codex docs: Codex Cloud (opens in a new tab), checked
  6. Codex docs: import from another agent (opens in a new tab), checked
  7. Codex docs: open source (opens in a new tab), checked
  8. ChatGPT docs: Sites (opens in a new tab), checked
  9. OpenAI API pricing (opens in a new tab), checked
  10. claude.com/pricing (opens in a new tab), checked
  11. claude.com/product/claude-code (opens in a new tab), checked
  12. Claude Code docs: overview (opens in a new tab), checked
  13. fabricate.build/pricing, from our live plans

About this guide

Written by the Fabricate team. Fabricate is our product; Codex and Claude Code are not, and the page marks ours where it mentions it.

The pages we read about each tool are listed here, each with the day we read it. We compared the tools from their own pricing pages and documentation, not by signing up for them. Spotted something out of date? Write to info@fabricate.build.

Or skip the comparison and build it

Describe it in a sentence and Fabricate builds it, backend included. Build free, go live on Pro.

Build free. No card needed.