What Is OpenAI Codex? The 2026 Coding Agent Explained
Last verified: June 16, 2026 · Format: Breakdown
OpenAI Codex is OpenAI's coding agent that runs in your terminal, your IDE, and the cloud. You hand it a software task in plain language, and it reads the relevant files, drafts the change, runs commands in your shell, and shows you a diff to approve. So if you are asking what is OpenAI Codex in practical terms, the short answer is this: one agent, three places to run it, and you stay in the review seat the whole way through.
Picture the everyday shape of it. You type "find where we validate signup emails, add a check for disposable domains, and write a test," and the tool does the legwork while you decide what actually lands in the repository. Where this article touches plans, it deliberately avoids quoting dollar figures, because OpenAI does not publish them in the sources behind this piece. Always confirm current details on OpenAI's Codex page before you commit. For where this assistant sits among other developer tools, see the AI Tools Hub.
What Is OpenAI Codex
OpenAI Codex is OpenAI's coding agent: a system you hand a software task to in plain language, and it searches your codebase, edits files, runs commands, and works through multi-step jobs on its own, pausing for you to review the result. The distinction worth drawing is between autocomplete and agency. Plenty of tools suggest the next line as you type; Codex aims higher, taking a whole task and returning a set of changes rather than a snippet to paste.
What makes Codex unusual is that it is not a single app. It is one agent exposed through three surfaces that share the same behavior: a command-line tool, an editor extension, and a web workspace. You can start a task at the keyboard in your terminal, continue it in your editor, and offload the long-running part of it to the cloud, all without switching to a different product or losing the thread.
Because Codex supports the Model Context Protocol (MCP), it can connect to external tools and data sources through a shared open standard rather than one-off integrations. It is built and maintained by OpenAI, the company behind ChatGPT, and the Codex CLI is openly developed on GitHub.
The Three Surfaces of Codex
Codex is one agent you can reach three ways. Each surface suits a different working style, and they are designed to interoperate rather than compete.
Codex CLI (terminal, open source)
The Codex CLI is the agent in your terminal. It is open source under the Apache-2.0 license and written in Rust, and it runs locally on your own machine. You launch an interactive text interface with the codex command, describe a task, and watch it read files, propose edits, and run commands. There is also a non-interactive mode for scripting and continuous integration, so the same agent can run unattended in a pipeline.
Codex IDE extension (your editor)
The Codex IDE extension brings the agent into your editor. It supports Visual Studio Code, the popular VS Code forks Cursor and Windsurf, and JetBrains IDEs. Inside the editor the agent can read your project, make edits in place, and let you review changes against your working tree, with the same approval controls as the CLI. On JetBrains, sign-in can also use a JetBrains AI subscription.
Codex Web and Cloud (the browser)
Codex Web is the cloud workspace, reached at chatgpt.com/codex. Here you can launch agent tasks that run on OpenAI's infrastructure rather than your laptop, which is useful for longer jobs you do not want tying up your local machine. The cloud is also where the local surfaces send their heavier work, which is the subject of the next section.
The practical takeaway: pick the surface that matches the moment. Terminal-first developers live in the CLI, editor-centric developers use the extension inside VS Code, Cursor, Windsurf, or JetBrains, and anyone who wants to fire off a long task from a browser uses Codex Web. The agent and your account follow you across all three.
How Local and Cloud Work Together
The most distinctive part of Codex is not any single surface but the loop between them. The local surfaces, the CLI and the IDE extension, connect to Codex Cloud so they can offload long-running jobs and then apply the results back to your machine.
In practice that means you can kick off a task that would take a while, a large refactor or a broad set of edits, and let it run in the cloud instead of blocking your editor. You monitor its progress without your local session freezing, and when it is done you preview the diff and apply the changes locally, on your terms. The cloud does the long work; you keep the final say over what lands in your repository.
From the CLI or the IDE extension, you hand a long-running job to the cloud rather than running it on your own machine.
The task runs on OpenAI's infrastructure. You can keep working locally and check on its progress without your session freezing.
You review the changes the cloud produced and apply them to your local repository only if you approve, keeping the final decision in human hands.
Core Features at a Glance
The agent is the headline, but Codex is several connected capabilities. Here is what each one does and why it matters day to day.
Approval modes
Codex gives you three levels of autonomy: Chat, Agent, and Agent (Full Access). In the most conservative mode the agent suggests and you drive; in the most permissive it can act more freely. The point is that you choose how much rope to give it for a given task, rather than accepting one fixed level of automation.
Subagents and parallel work
Codex can spin up subagents to handle parts of a job in parallel, so a larger task can be broken into pieces that progress at the same time. This is how it tackles work that has several independent strands without forcing everything through a single sequential thread.
Code review, MCP, and web search
Codex includes a local code-review agent that can inspect changes before you commit them, a useful second pair of eyes on a diff. It supports the Model Context Protocol to connect third-party tools and data sources through one open standard, and it can perform web search when a task needs current information beyond the codebase.
Images in and out, and scripting
Codex accepts image inputs, so you can hand it a screenshot or a mockup as part of a request, and it can produce image output as well. For automation, the non-interactive exec mode lets you script the agent into pipelines and CI, where it runs without prompting for approval.
Which Models Codex Uses
Codex runs on OpenAI's GPT-5 series. The latest model it references is GPT-5.5, with GPT-5.4 also available. You switch between them with the /model command, and you can set the reasoning effort to low, medium, or high depending on how much deliberation a task warrants. Lower effort is faster; higher effort spends more time reasoning through a hard problem.
| Setting | Options (as of June 16, 2026) |
|---|---|
| Model | GPT-5.5 (latest), GPT-5.4 |
| Switch models | /model command |
| Reasoning effort | Low, Medium, High |
A naming note: the sources behind this article reference GPT-5.5 and GPT-5.4 only. You may see write-ups elsewhere name a dedicated Codex-branded model variant; we do not assert any such model here, because it is not in the documentation we verified. Treat the model roster as fast-moving and confirm the current names in OpenAI's Codex documentation.
How to Install Codex and What Plans Include It
Getting the CLI is a one-liner on macOS and Linux, and there are several ways in depending on your platform and tooling.
Installing the CLI
- macOS / Linux: run curl -fsSL https://chatgpt.com/codex/install.sh | sh. For unattended installs, set CODEX_NON_INTERACTIVE=1.
- Windows: Codex runs natively in PowerShell with a Windows sandbox, or you can use WSL2.
- Package managers: it is also available via npm and Homebrew, or as prebuilt binaries from GitHub Releases.
- IDE extension: install it from the VS Code Marketplace, or from the JetBrains plugin marketplace.
Signing in
The first time you run Codex it prompts you to sign in. You can authenticate with your ChatGPT account, which draws on your plan's usage credits, or with an API key. On JetBrains, a JetBrains AI subscription is also supported.
Which plans include Codex
Codex is included in ChatGPT Plus, Pro, Business, Edu, and Enterprise plans. The sources behind this article do not list specific dollar amounts or numeric usage limits for those plans, so rather than quote figures that could be wrong, we point you to the official pages. Confirm current pricing and rate limits on OpenAI's Codex page and the Codex web workspace before you rely on any specific number.
- Open-source Rust CLI, runs locally
- Sign in with ChatGPT account or API key
- Interactive TUI plus non-interactive exec
- Connects to Codex Cloud for long jobs
- Codex included in the subscription
- Uses plan usage credits
- CLI, IDE extension, and Codex Web
- Figures not quoted here, confirm at openai.com
- Codex included for organizations
- Shared workspace administration
- All three surfaces available
- Confirm seat terms with OpenAI
- Codex included at the enterprise tier
- Organization-level controls
- API-key auth also supported
- Pricing and limits via sales
Why no dollar figures? We only publish numbers we can ground in a verified source. The documentation behind this article confirms which ChatGPT plans include Codex but does not state prices or numeric rate limits, so quoting a figure here would be guesswork. Check the live pricing page for the current amounts.
Who Codex Is For
By now the answer to what is OpenAI Codex should be clear: it is a coding agent that meets you where you already work. It fits a wide range of developers, but the surface that suits you best depends on how you work. Here is how the main groups line up.
People who live in the shell and want the agent where they already work. The open-source CLI installs in seconds and runs locally, with a scripting mode for automation.
Best fit: Codex CLIDevelopers who prefer working inside an editor. The IDE extension runs in VS Code, the Cursor and Windsurf forks, and JetBrains IDEs, with in-place edits and diff review.
Best fit: Codex IDE extensionPeople who want to fire off a large refactor and not tie up their machine. Codex Web launches cloud tasks you can monitor without blocking, then apply locally.
Best fit: Codex Web / CloudOrganizations already on ChatGPT Business, Edu, or Enterprise, where Codex is included. API-key authentication supports automation and shared workflows.
Best fit: Business / EnterpriseHonest Limitations
Codex is a capable tool, and the points below are not reasons to avoid it. They are reasons to use it with clear eyes.
An agent that edits files and runs commands can move fast in the wrong direction just as easily as the right one. The approval modes and local code-review agent exist for a reason: read the changes before you accept them, especially in Agent (Full Access) mode where it acts more freely.
Codex is included in several ChatGPT plans, but the sources behind this article do not state dollar figures or numeric usage limits. Do not assume a specific price or rate cap from any single write-up, including this one. Check OpenAI's pricing page for the current terms before you budget.
Offloading a job to Codex Cloud means your code and context are processed on OpenAI's infrastructure rather than locally. For proprietary or regulated codebases, review OpenAI's data-handling terms and your organization's policy before delegating sensitive work to the cloud.
Model names, surfaces, and capabilities change quickly here. The GPT-5 series has iterated rapidly, and feature sets shift between releases. Anchor decisions to the live OpenAI documentation rather than to any single snapshot, including this one.
Frequently Asked Questions
Go Deeper
Resources from across Tech Jacks Solutions
AI Tools Hub
Browse every AI tool we have reviewed, by vendor and use case
AI Glossary
Plain-language definitions for AI and developer terms used here
FREEAI Governance Charter
Set rules for AI coding tools before they scale across your team
EU AI Act Overview
How AI regulation frames code generation and developer tooling