If you code every day and are deciding between OpenAI Codex and Anthropic’s Claude Code, the official docs point to a practical answer. The two tools overlap heavily, so the account you already pay for and the places you work should decide it.
Anthropic calls Claude Code “an agentic coding tool”: it reads your codebase, edits files, runs commands and connects to your development tools. OpenAI’s Codex CLI page doesn’t use the word “agentic”, but it describes the same kind of work: Codex inspects code, makes changes, runs commands and automates repeatable work from the terminal. Both tools run in a terminal, in an editor and in the cloud. Both document MCP (Model Context Protocol, an open standard for connecting agents to outside tools and data), subagents, a non-interactive mode for scripts and CI, controls over what the agent may do, and automated pull-request review on GitHub.
The biggest difference is how you get access. Codex is included in every ChatGPT plan from Free to Enterprise, and it can also run on an OpenAI API key. Plan level still matters. The Free and Go cards list only desktop-app access. Cloud integrations such as GitHub code review and Slack are listed for Plus, Pro, Business and Enterprise/Edu, and not for the API-key route. Most Claude Code surfaces need a Claude subscription or an Anthropic Console account. Its terminal, VS Code and JetBrains versions can also run through third-party providers.
The workflows differ in a few places too. For example, Anthropic documents cloud Routines that keep running when your computer is off, while OpenAI’s feature table lists Codex’s scheduled tasks under local features. Neither vendor publishes head-to-head quality or speed data, and no benchmark was run for this guide, so it doesn’t name a winner. Below you’ll find a side-by-side table built from the official pages, install steps, plan details, the workflow differences, and a short “choose X if…” list.
Codex vs Claude Code at a glance
This table compares what each vendor documents. It doesn’t rate quality or speed: neither vendor’s docs include comparative data, and no tests were run for this guide. Plans, prices and models change often, so treat it as a snapshot of the pages on the date shown.
OpenAI Codex vs Claude Code, from official docs as checked on September 27, 2026
| | OpenAI Codex | Claude Code |
| — | — | — |
| What it is | OpenAI’s CLI page says Codex inspects code, makes changes, runs commands, and automates repeatable work | Anthropic calls it an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with dev tools |
| Where it runs | CLI, IDE extension, web, iOS (listed on Plus), desktop app, and Codex cloud | Terminal CLI, VS Code extension (with a Cursor install link), JetBrains plugin, desktop app, web at claude.ai/code, and Claude mobile apps |
| How you get access | Included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise; or use an OpenAI API key | A Claude subscription or Anthropic Console account on most surfaces; the terminal CLI, VS Code, and JetBrains also support third-party providers |
| Listed prices | Free $0; Go $8/mo; Plus $20/mo; Pro from $100/mo; Business $20/user/mo billed annually or $25 billed monthly (2+ users); Enterprise/Edu through sales | Not listed in the Claude Code docs; see Anthropic’s pricing page |
| Project instructions file | AGENTS.md (created with /init) | CLAUDE.md; can also read an existing AGENTS.md |
| Scripting and CI | codex exec; Codex SDK | claude -p; GitHub Actions and GitLab CI/CD integrations; Agent SDK |
| Code review | /review on uncommitted changes, a commit, or a base branch; GitHub code review and automatic PR reviews on Plus, Pro, Business, and Enterprise/Edu (not with an API key) | Review automation through GitHub Actions or GitLab CI/CD; GitHub Code Review for automatic review on every PR |
| Permissions and hooks | /permissions sets what runs without asking and shows the active sandbox (the limits on what the agent can reach); the docs also have Hooks, Sandboxing, and Auto-review pages | Hooks run shell commands before or after actions; the changelog covers /permissions rules and sandbox settings |
| MCP | codex mcp adds local or remote servers | Supported; MCP servers work across all surfaces |
| Subagents | Delegates focused work to specialized agents; feature table lists subagents and custom agents | Spawns parallel agents; background agents; Agent SDK for custom agents |
| Cloud and scheduled work | codex cloud submits work to a configured environment and applies results locally; feature table lists scheduled tasks under local features | Web sessions; cloud Routines that keep running with your computer off; desktop scheduled tasks; /loop |
| Team chat | Slack cloud integration on Plus, Pro, Business, and Enterprise/Edu (not with an API key) | Mention @Claude in Slack with a bug report and get a pull request back |
Claude Code overview (Anthropic)
Where each tool runs
Anthropic’s overview lists five Claude Code surfaces. The terminal CLI is the full-featured version. The VS Code extension adds inline diffs, @-mentions, plan review, and conversation history, and the same page has a one-click install link for Cursor. The JetBrains plugin works in IntelliJ IDEA, PyCharm, WebStorm, and other JetBrains IDEs, but you have to install the CLI separately first. The desktop app includes Claude Code, so you don’t need a separate CLI, but it requires a paid subscription. The web version at claude.ai/code needs no local setup. It’s meant for long-running or parallel tasks and for repositories you don’t have on your machine. Anthropic says every surface runs on the same engine, so your CLAUDE.md files, settings, and MCP servers work the same everywhere.
On OpenAI’s pricing page, the Plus plan lists Codex on the web, in the CLI, in the IDE extension, and on iOS. Pro includes everything in Plus. The Business card says it gives access to ChatGPT and Codex across desktop and mobile apps. The Free and Go plans each list one line: GPT-6 Luna at Standard speed in the desktop app, subject to rollout. OpenAI’s feature table also lists mobile remote control and SSH remote connections.
The Codex CLI page documents codex cloud, which sends work to a configured cloud environment and pulls the result back into your local repository. The CLI and pricing pages confirm that a Codex IDE extension exists, but they don’t say which editors it supports. If that matters to you, check OpenAI’s IDE extension page.
Installing and running your first task
Both tools publish a one-line installer for macOS and Linux. Claude Code’s native installer also works in WSL. For native Windows, Anthropic documents a PowerShell installer (irm https://claude.ai/install.ps1 | iex), a CMD installer, and winget install Anthropic.ClaudeCode. Homebrew users can run brew install --cask claude-code, and Debian, Fedora, RHEL, and Alpine users can install with apt, dnf, or apk. Homebrew offers two casks. claude-code follows the stable channel, which Anthropic says typically runs about a week behind. claude-code@latest gets new versions as they ship.
Native installs update themselves in the background. Homebrew and WinGet installs don’t. Update them yourself with brew upgrade claude-code (or brew upgrade claude-code@latest, if that’s the cask you installed) or winget upgrade Anthropic.ClaudeCode.
On first launch, Claude Code asks you to log in. If the ANTHROPIC_API_KEY environment variable is set, it skips the login and asks you to approve that key instead. Codex asks you to choose Sign in with ChatGPT or another available sign-in method. OpenAI’s install section also has Windows, npm, and Homebrew tabs. This guide doesn’t reproduce those commands, so use the tabs on the Codex CLI page.
OpenAI recommends making Git checkpoints before and after each task so you can revert changes. That advice works just as well for Claude Code.
# Claude Code (macOS, Linux, WSL)
curl -fsSL https://claude.ai/install.sh | bash
cd your-project
claude
# Codex CLI (macOS, Linux)
curl -fsSL https://chatgpt.com/codex/install.sh | sh
cd your-project
codexWhat plan or account gets you access
OpenAI’s pricing page says Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise. These were the listed prices on September 27, 2026:
- Free: $0
- Go: $8/month
- Plus: $20/month, which includes cloud integrations such as automatic code review and Slack
- Pro: from $100/month, with a choice of 5x or 20x Plus’s usage
- Business: $20 per user per month billed annually, or $25 billed monthly, with a two-user minimum
- Enterprise and Edu: through sales
You can also run Codex on an API key, in the CLI, SDK, or IDE extension. You pay API rates, and the models you get depend on your key. This route doesn’t include cloud features such as GitHub code review or Slack. OpenAI says it’s a good fit for automation in shared environments such as CI.
OpenAI gives Codex usage limits as estimated local messages per five-hour window. The estimates vary by model and task size, and OpenAI says they aren’t fixed limits. Local and cloud work share the same allowance, and weekly limits may also apply. During a CLI session, /status shows how much you have left.
Reviews that run through GitHub count as Code Review usage. That covers tagging @Codex on a pull request or turning on automatic reviews. Local reviews count toward your general limits. When Plus and Pro users hit a limit, they can buy extra credits. Any user can also keep running local chats on an API key at standard API rates. Two more notes from the pricing page: image generation isn’t available on the Free plan, and GPT-5.5 will be retired from Codex on all plans on October 14, 2026.
The Claude Code docs describe access requirements but list no prices. Most surfaces need a Claude subscription or an Anthropic Console account. The desktop app needs a paid subscription. Pulling a web session into your terminal with claude --teleport requires a claude.ai subscription. Changelog entries mention the Pro, Max, Team, and Enterprise plans. They also say Free-plan users who open claude.ai/code are sent to an upgrade page, and that self-hosted runners are available on Team and Enterprise. For Claude prices and usage limits, check Anthropic’s pricing page.
How the workflows actually differ
Project instructions carry over between the two tools. Codex’s /init command creates an AGENTS.md file. Claude Code reads CLAUDE.md, and Anthropic says it can also read an existing AGENTS.md, either alone or alongside CLAUDE.md. A changelog entry adds that in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead. That means a team using both tools can keep a single shared instructions file.
Both tools can run inside scripts. Claude Code follows the Unix model: you can pipe logs or a git diff into claude -p and chain it with other tools. Codex offers codex exec for the same kind of non-interactive run, plus a Codex SDK.
Both also handle review at two levels. Codex’s /review checks uncommitted changes, a single commit, or a comparison against a base branch. It reports prioritized findings without touching your working tree. On Plus, Pro, Business, and Enterprise/Edu plans, Codex can also review pull requests on GitHub, either when tagged or automatically; the API-key route doesn’t include this. Claude Code documents review through GitHub Actions or GitLab CI/CD, plus a GitHub Code Review integration that reviews every PR.
Both tools let you control what the agent may do on its own. A key idea here is the sandbox: the limits on which files, folders, and commands the agent can reach without asking you first. Codex’s /permissions sets when Codex can edit files or run commands without asking. It also shows the active sandbox and the folders Codex can write to. OpenAI’s docs have separate pages on hooks, sandboxing, and auto-review. Claude Code documents hooks, which are shell commands that run before or after its actions, such as auto-formatting after every edit. Its changelog also covers /permissions rules and sandbox settings. Neither vendor compares its controls with the other’s, so treat this as shared ground rather than a reason to pick one tool.
For larger jobs, both tools document subagents: separate agents that each take a slice of the work and report back. Codex’s feature table lists subagents and custom agents. Claude Code’s overview describes parallel agents coordinated by a lead agent. It also names background agents you can watch from one screen and an Agent SDK for building your own agents. On OpenAI’s side, the matching build-your-own option is the Codex SDK.
The clearest documented difference is scheduling. Anthropic says Claude Code Routines run in the cloud, keep running when your computer is off, and can trigger on API calls or GitHub events. OpenAI’s feature table lists Codex’s project and standalone scheduled tasks under local features. That grouping is the only Codex scheduling detail in the pages read for this guide. OpenAI’s docs have a separate Scheduled tasks page, which wasn’t read here, so check it before assuming Codex can’t schedule cloud work.
For moving work between devices, Claude Code documents Remote Control, claude --teleport, /desktop, and claude --cloud. Codex offers codex cloud from the terminal, web and iOS access on Plus, and mobile remote control in its feature table.
Team chat works differently in each tool. Codex has a Slack cloud integration on Plus, Pro, Business, and Enterprise/Edu plans. Claude Code lets you mention @Claude in Slack with a bug report and get a pull request back.
The Codex CLI page also lists a few everyday options. codex resume reopens saved chats, --image attaches screenshots or diagrams, --search turns on live web search, and /model picks the model and reasoning effort.
Choose Codex if… choose Claude Code if…
These suggestions match each tool to a setup. They aren’t a verdict on which one writes better code.
- Choose Codex if you already pay for ChatGPT Plus or Pro and want Codex on the web, in the CLI, in the IDE extension, and on iOS on that plan. Business lists Codex across desktop and mobile apps; check OpenAI’s feature table for the rest.
- Choose Codex if you want to try it for free. Keep in mind that Free and Go list access through the desktop app, subject to rollout.
- Choose Codex if you want to run it in CI on an OpenAI API key and don’t need cloud features such as GitHub code review or Slack.
- Choose Claude Code if you already have a Claude subscription or Anthropic Console account. If you need a third-party provider, Claude Code supports one on its terminal CLI, VS Code, and JetBrains surfaces. For Codex, see OpenAI’s Amazon Bedrock page under deployment and model providers.
- Choose Claude Code if you want scheduled jobs that run in the cloud and keep going with your laptop closed. Anthropic documents that for Routines. OpenAI’s feature table lists Codex’s scheduled tasks as a local feature, but read OpenAI’s Scheduled tasks page before ruling Codex out.
- If you work in VS Code, Cursor, or a JetBrains IDE, Claude Code documents an extension or plugin for each. Codex has an IDE extension too; check OpenAI’s IDE extension page for supported editors before you decide.
- On a mixed team, keep shared rules in AGENTS.md and make a Git commit before and after every agent task, whichever tool runs it.
What the docs and this guide leave open
Neither vendor publishes head-to-head accuracy or speed data, so this guide doesn’t cover which tool is smarter. Anthropic’s Claude Code docs don’t list subscription prices or usage limits; those are on Anthropic’s pricing page. OpenAI’s CLI and pricing pages don’t say which editors the Codex IDE extension supports, so check the IDE extension page. This guide also doesn’t reproduce the Codex CLI’s Windows, npm, and Homebrew install commands; they’re on the install tabs of the Codex CLI page. The changelog text read for this guide shows no version or date headers, so it can’t tell you when a given Claude Code feature shipped. Both vendors change plans, limits, and default models often, so check the pricing pages again before you commit.
Start with the account you already have. If you pay for ChatGPT Plus or higher, install the Codex CLI and sign in with ChatGPT. If you have a Claude subscription or an Anthropic Console account, install Claude Code and run claude in a project. If you want to try both, put shared project rules in an AGENTS.md file, since both tools can read one. Make a Git commit before and after each agent task so you can roll back. If your editor matters, check OpenAI’s Codex IDE extension page for the editors it supports. If you need scheduled jobs, read OpenAI’s Scheduled tasks page before assuming Codex runs them only locally. Before you buy, check both vendors’ pricing pages again, because plans and models change often.