Back
OpenAI CodexTRY IT

OpenAI Codex

OpenAI Codex is OpenAI's coding agent, included with ChatGPT plans. It reads a codebase, edits files, runs commands and tests, and hands back changes for you to review, whether you're working in the ChatGPT desktop app, the terminal, a code editor or the cloud. All of these are tied to one ChatGPT account, so a task started in one place can be picked up in another.

When this listing was first written, Codex was a cloud agent in ChatGPT that ran tasks in sandboxes and opened pull requests. That part still exists as Codex Cloud. Since then OpenAI has added an open-source command-line tool, editor extensions, a desktop app, automatic pull request reviews and plugins, and Codex is now as much a local assistant as a background one.

The Codex desktop app: a sidebar of recent tasks and projects on the left and a "What should we build?" prompt box in the main window, with the model picker set to 6 Astra High

The Codex side of the ChatGPT desktop app, with recent tasks grouped by project. Screenshot from OpenAI.

Where you can use Codex

The ChatGPT desktop app for macOS, Windows and Linux has a Codex mode next to the usual chat. You open a project folder, describe the task and follow the agent's progress, and several tasks can run in parallel. The app can also use a browser and other desktop apps to check its own work.

Codex CLI runs in your terminal against a local repository. It's open source under the Apache 2.0 license, installs with a script, npm or Homebrew, and can be called from scripts and CI pipelines with codex exec. The IDE extension works in VS Code, Cursor and Windsurf, and Xcode and JetBrains IDEs have their own Codex integrations. Both let you pick the model, reasoning effort and permissions for each task.

Codex Cloud runs tasks in isolated workspaces built from a reusable environment. You choose GitHub repositories, Codex installs dependencies and tests the setup, and tasks keep going while your computer sleeps. You can review the results from the web, the desktop app or the mobile app, then commit or open a pull request.

Code review and team integrations

Codex can review GitHub pull requests automatically or when someone mentions @codex review, and it answers questions about the findings. GitLab merge request reviews are in preview. A Slack integration lets teams hand Codex tasks from a channel, and a Security Cloud plugin, in research preview, scans connected repositories for vulnerabilities and proposes patches.

Control and customization

Local work runs in a sandbox with approval rules: you decide whether Codex can only read, can write inside the workspace, or can run commands and reach the network. An AGENTS.md file in a repository gives it project instructions. Skills, plugins, MCP servers and subagents extend what it can do, and the Codex SDK and app server let developers build it into their own tools.

Plans and pricing

Codex usage comes out of your ChatGPT plan and is shared with ChatGPT Work. Limits depend on the model, the size of the task and whether it runs locally or in the cloud.

PlanWhat you get
Free and GoCodex in the desktop app with a lighter model, for quick tasks, subject to rollout
PlusCodex on the web, in the CLI, in the IDE extension and on iOS, plus code review, Slack integration and the latest models
ProHigher usage limits in three tiers, with the fastest model mode on the top tier
BusinessA shared workspace with admin controls, SAML SSO, larger cloud machines and no training on business data by default
Enterprise and EduPriority processing, SCIM, role-based access, audit logs, and data retention and residency controls
API keyThe CLI, SDK and IDE extension billed at API rates, without cloud features such as code review or Slack

Extra usage can be bought as ChatGPT credits on most paid plans.

Is OpenAI Codex free?

Free and Go accounts can try Codex in the desktop app with a lighter model as it rolls out. The CLI is open source, but it still needs a ChatGPT plan or an API key to run models. Cloud tasks, code review and Slack start on the Plus plan.

What's the difference between Codex CLI and Codex Cloud?

The CLI works on your own machine, in your local repository, with the tools you already have installed. Codex Cloud runs each task in a remote workspace connected to GitHub, so it can work in parallel and keep going when your laptop is closed.

Can Codex review pull requests?

Yes. Once GitHub is connected, Codex can review pull requests automatically or on request, and the desktop app has a Code Review view for reading diffs, comments and checks in one place.

Is OpenAI Codex the same as the old Codex model?

No. The original Codex was a 2021 code-generation model behind early GitHub Copilot, and it was retired. Today's Codex is an agent product that runs on OpenAI's current GPT models.

Alternatives to OpenAI Codex

  • Claude: Anthropic's assistant, whose Claude Code agent works in the terminal, IDEs, a desktop app and the web.
  • GitHub Copilot: Code completions, chat and an agent mode inside VS Code and other editors, with a cloud agent that works on GitHub issues.
  • Cursor: AI-first code editor built on VS Code, with tab completion and an autonomous agent mode.
  • Google Antigravity: Google's desktop agent app for coding and project work, with parallel subagents, scheduled tasks and visual review.
  • Jules: Google's asynchronous coding agent that connects to GitHub and handles routine tasks in the background.
  • Windsurf: AI-native code editor whose Cascade agent edits across the whole codebase.
  • Cline: Free, open-source coding agent for VS Code, JetBrains and the terminal that works with the model of your choice.
  • OpenCode: Open-source coding agent with terminal, desktop and IDE interfaces and support for many model providers.