What is Codex?
Codex is OpenAI’s AI coding agent that reads your repository, writes and runs code, and returns reviewable changes such as bug fixes, refactors and pull requests. It works through your ChatGPT account rather than a separate subscription.
You describe a task in plain language, and Codex plans the steps, edits files and runs commands or tests in a sandbox. It can work locally through the CLI, the IDE extension and the ChatGPT desktop app, or remotely in Codex cloud, where environments persist so a job can continue while you are away. You then inspect the diff and decide what to merge.
OpenAI builds and runs the service, and the CLI is open source on GitHub. The name causes confusion, because OpenAI used it in 2021 for a GPT-3-based code model that formed the basis of GitHub Copilot. That older model was discontinued in March 2023, and today’s Codex is a different, agent-based product.
Key Features
Cloud tasks and the open-source CLI matter most, because together they let Codex work both in the background on remote environments and directly on your own machine.
- Codex cloud: Codex cloud runs tasks in remote environments linked to your repositories, so work continues after you close the laptop. You can queue several jobs, review the diffs later and open pull requests from the results.
- Codex CLI: The open-source CLI runs as a local agent in your terminal and can read, edit and run code in the current folder. It is released under the Apache-2.0 licence and installs through npm, Homebrew or a script.
- IDE extension: An extension brings the agent into VS Code, Cursor and Windsurf, according to the GitHub project page. It uses the same ChatGPT account as the other surfaces, so one plan covers work in the editor, the terminal and the cloud.
- Parallel agents: Codex can split a larger job across several agents working at the same time and then check the results. Directories describe this as worktree-based, which keeps each attempt separate until you choose one to merge.
- Automated code review: Codex can review pull requests on GitHub and flag likely bugs before a human reviewer opens them. Teams use this as an extra pass, not as a replacement for their normal approval rules.
- Plugins and MCP: Plugins connect Codex to GitHub, Slack, Gmail, Google Drive, Figma, Notion and Supabase, according to the product page. The documentation also covers Model Context Protocol servers and AGENTS.md files for project instructions.
Pros
Bundled access is the strongest draw: anyone on ChatGPT Plus at $20 a month already has Codex across the web, CLI, IDE and iOS.
- One account, four surfaces: A single ChatGPT login covers the desktop app, the web, the CLI and the IDE extension. You can start a task in the terminal and check it later on another device without extra setup.
- Included in existing plans: Codex usage comes bundled with ChatGPT Plus, Pro and Business rather than sold as a separate product. Developers who already pay $20 a month for Plus can try agentic coding at no extra cost.
- Open-source CLI: The CLI code is public on GitHub under Apache-2.0, where the project shows more than 127,000 stars. Security teams can inspect how the local agent handles files, commands and approvals.
- Handles long tasks: Background cloud environments let Codex work on refactors, migrations or test fixes while you do something else. This suits multi-file changes that would take a chat assistant many rounds of copying.
- Controlled autonomy: The docs describe sandboxing, permission settings and approval workflows for commands the agent wants to run. You decide how much the agent may do on your machine before it asks.
Cons
Usage limits are the catch with Codex, since Plus allows 350 to 3,000 local messages per five hours and heavy users may need $100 or more.
- Usage caps vary: Plus quotes a wide range of 350 to 3,000 local messages per five hours, depending on model and task size. Heavy users can hit the ceiling mid-session, so the Pro tiers are often needed.
- Top tiers are expensive: Pro starts at $100 a month and rises to $200 and $500 tiers. Freelancers who need the larger limits or the fastest model face a cost well above most rival coding assistants.
- Free access is narrow: Free and Go users get only the lighter GPT-6 Luna model in the desktop app, and OpenAI says this is subject to rollout. The CLI, IDE extension and cloud features need Plus or higher.
- Output still needs review: Directories note that Codex can write redundant code or misread complex intent. Every change should pass tests and a human review before it reaches your main branch.
- Code leaves your machine: Cloud tasks send repository contents to OpenAI servers for processing. Teams with strict data rules should check Business or Enterprise terms, which state no training on business data by default.
Who is Using Codex?
Professional developers form the core audience, alongside engineering teams that want an agent for refactors, migrations and pull request review.
- Software engineers: Developers hand Codex bug fixes, refactors and test writing. They review the resulting diffs and merge the changes they accept.
- Engineering leads: Tech leads use automated pull request review as an extra check. It helps them catch issues before senior reviewers spend time on a change.
- Platform teams: Platform and reliability engineers use it for migrations and CI fixes across many files. Parallel agents let them try several approaches at once.
- Students and hobbyists: Learners use the CLI or the desktop app to explore unfamiliar codebases. Free and Go users can try a limited version before paying for Plus.
Pricing
Codex is freemium, with limited free access and full use from $20 a month on ChatGPT Plus (official pricing page, October 2026).
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | GPT-6 Luna at standard speed in the desktop app, subject to rollout |
| Go | $8 per month | Same Luna access in the desktop app, subject to rollout |
| Plus | $20 per month | Web, CLI, IDE extension and iOS; GPT-6.1 Sol and GPT-6 Luna; 350 to 3,000 local messages per five hours |
| Pro | $100, $200 or $500 per month | No fixed five-hour limit on local messages; Astra Ultrafast on the $500 tier |
| Business | $20 per user per month (annual) or $25 monthly | Shared workspace, SAML SSO, MFA and no training on business data by default |
| Enterprise and Edu | Contact sales | SCIM, EKM, RBAC, audit logs and flexible credit-based pricing |
| API key | Standard API token rates | Pay per token instead of using a ChatGPT plan |
Futurepedia lists Codex as freemium from $8 a month, which matches the Go plan but understates how limited that tier is for coding work.
Disclaimer: OpenAI adjusts plan limits and model access often, so confirm the current figures on its pricing page before you subscribe.
What Makes Codex Unique?
Persistent cloud environments tied to a ChatGPT login set Codex apart. Many coding assistants work only inside one editor session, whereas Codex can keep a task running remotely, split it across parallel agents and let you pick it up later from the app, the web or the terminal.
The second difference is openness at the local layer. The CLI source is public under Apache-2.0, and the agent reads project rules from AGENTS.md files and connects to tools through MCP. That combination gives teams more visibility and control than a closed editor plugin usually offers.
Compatibilities and Integrations
Codex runs on macOS, Linux and Windows, inside popular editors, and connects to GitHub, Slack and other services through plugins.
- Platforms: The CLI supports macOS and Linux, with a PowerShell installer and WSL guidance for Windows. Codex cloud runs on the web, and Plus includes iOS access.
- Editors: The IDE extension works with VS Code, Cursor and Windsurf. The same ChatGPT plan covers editor use.
- Services: Plugins cover GitHub, Slack, Gmail, Google Calendar, Google Drive, Figma, Notion and Supabase. GitHub also handles automated pull request review.
- Extensibility: MCP servers add custom tools, and AGENTS.md files store project instructions. Sign-in works with a ChatGPT plan or an API key.
Codex Tutorials
Setup takes a few minutes with a ChatGPT Plus account and the CLI or IDE extension.
- Sign in to ChatGPT with a Plus, Pro or Business plan, or prepare an API key.
- Install the CLI with npm install -g @openai/codex or brew install –cask codex, or add the IDE extension.
- Open your project folder, run codex and sign in when prompted.
- Add an AGENTS.md file describing your build commands, tests and coding style.
- Describe a small task, such as fixing a failing test, and approve commands as they appear.
- Review the diff, run your tests and only then commit or open a pull request.
How We Rated It
Codex earns 4.2 out of 5 thanks to its breadth across cloud, CLI and IDE, while usage caps and high top-tier prices pull the score down.
| Criterion | Score | Why |
|---|---|---|
| Accuracy and Reliability | 4.3/5 | Strong fixes, still needs review |
| Ease of Use | 4.0/5 | Simple app, CLI learning curve |
| Functionality and Features | 4.7/5 | Cloud, CLI, IDE, review |
| Performance and Speed | 4.2/5 | Parallel agents, variable waits |
| Customization and Flexibility | 4.4/5 | AGENTS.md, skills, MCP |
| Data Privacy and Security | 3.9/5 | Sandboxing, code leaves machine |
| Support and Resources | 4.0/5 | Good docs, limited human help |
| Cost-Efficiency | 3.6/5 | Usage limits, costly top tiers |
| Integration Capabilities | 4.5/5 | GitHub, Slack, plugins, API |
| Overall Score | 4.2/5 | Average of the nine criteria |
Summary
Codex suits developers and engineering teams that already pay for ChatGPT and want an agent able to handle multi-file changes, pull request review and background tasks. Its main limitation is usage, because the Plus allowance can run out during heavy sessions and the larger Pro tiers cost $100 to $500 a month. Free and Go access is too narrow to judge the product fairly. Like any coding agent, it still needs tests and human review before code reaches production. Start with one well-defined task and a clear AGENTS.md file, then widen its scope once you trust the output.
Reviews
Write a Review