Pick Claude Code if you already pay for Claude and want the deepest extension layer: 33 hook events, skills, plugins and cloud routines that fire on a schedule or a GitHub event. Pick Codex if you already pay for ChatGPT: it comes with every ChatGPT plan, its CLI is open source, and you can hand tasks to the cloud from ChatGPT, GitHub or Slack. The two now share most building blocks (AGENTS.md, skills, MCP, hooks, subagents), so the real choice is model family, plan and where you like to work.
Codex and Claude Code at a glance
Prices and features below are as of October 2026, from each vendor's own pricing and docs pages.
| Claude Code | Codex | |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Price | Pro $20/month ($17 billed annually), Max $100 or $200, Team $25 per seat ($125 for a Premium seat), Enterprise $20 per seat billed annually plus usage; or an API key | In every ChatGPT plan: Free and Go ($8) in the desktop app only, Plus $20, Pro $100, $200 or $500, Business $25 per user ($20 billed annually); or an API key |
| Models | Claude only: Opus 5.5 by default, Sonnet 5.5, Haiku 5.5, Fable 5.1 on request | OpenAI only: the GPT-6 family on ChatGPT plans |
| Where it runs | Terminal, VS Code, JetBrains, desktop app, web, Claude mobile app, Slack | CLI, IDE extension (VS Code, Cursor, Windsurf), ChatGPT desktop app, web, mobile, Slack, Linear |
| Cloud agents | Cloud sessions on the web, background agents, Projects | Codex cloud tasks from the web, @codex on GitHub, Slack |
| Project instructions | CLAUDE.md (can read AGENTS.md instead) | AGENTS.md |
| Skills | SKILL.md, Agent Skills standard; custom commands merged into skills | SKILL.md in ~/.agents/skills or .agents/skills |
| MCP | Yes | Yes |
| Hooks | 33 events; command, HTTP, MCP tool, prompt or agent handlers | 12 events; each hook must be trusted with /hooks |
| Scheduling | Cloud routines (hourly at most), desktop scheduled tasks, /loop | Scheduled tasks (automations) |
| Platforms | macOS, Linux, Windows | macOS, Linux, Windows |
Pricing: a Claude plan or a ChatGPT plan
Both are sold mainly through a chat subscription, and both meter use in five-hour windows rather than by message count.
- Claude Code is not in the free Claude plan. Pro ($20 a month) gives at least five times the free plan's usage per five-hour session; Max gives five or twenty times Pro. Paid plans also have weekly limits. Fable, Anthropic's largest model, is never the default and can bill to usage credits instead of your plan.
- Codex is in every ChatGPT plan. Free and Go users get it in the ChatGPT desktop app with GPT-6 Luna. Plus adds the CLI, the IDE extension, Codex on the web and integrations such as automatic code review. OpenAI publishes estimated local messages per five hours for Plus (for example 350 to 3,000 with GPT-6 Luna, 15 to 160 with GPT-6.1 Sol) and says these are not fixed limits. Pro has no five-hour limit for now.
Both also work with a pay-per-token API key. With Codex, an API key covers the CLI, SDK and IDE extension but none of the cloud features.
Where you work
Claude Code is terminal-first, but every surface runs the same engine, so your CLAUDE.md, settings and MCP servers follow you. You can start a task on the web or the phone app and pull it into your terminal with claude --teleport, move a terminal session into the desktop app with /desktop, or keep driving a local session from your phone with Remote Control. In the desktop app, Claude Code is the Code tab.
Codex changed shape in July 2026, when OpenAI merged the Codex app into the new ChatGPT desktop app. Codex is now a mode there, next to chat and ChatGPT Work. The Codex CLI is still its own open source project (Apache-2.0) that you install with npm, Homebrew or a script, and Codex on the web runs tasks in OpenAI's cloud.
Customization: almost the same building blocks
The two agents have converged:
- Instructions: Codex reads
AGENTS.md. Claude Code readsCLAUDE.md, and can read an existingAGENTS.mdin its place. - Skills: both use a folder with a
SKILL.mdfile. Claude Code follows the open Agent Skills standard and adds its own fields; a skill that only uses the standard fields (name,description,license,compatibility,metadata,allowed-tools) is portable. - Hooks: Claude Code fires hooks on 33 lifecycle events, from
PreToolUseandPermissionRequesttoStopandSessionEnd, and a hook can be a shell command, an HTTP call, an MCP tool call, a prompt or a subagent. Codex has 12 events in~/.codex/hooks.json(orconfig.toml), and every non-managed hook has to be reviewed and trusted with/hooks; change it and it is skipped until you trust it again. See our Claude Code hooks guide for examples. - MCP, subagents and plugins: both have all three.
Running unattended
For a quick non-interactive run, both have a print mode: claude -p "summarize the failing tests" and codex exec "summarize the failing tests". Both fit in CI and cron.
For longer work, Claude Code has cloud sessions on the web, background agents you watch from one screen, subagents, and routines: a saved prompt plus repositories and connectors that runs in Anthropic's cloud on a schedule, from an HTTP call, or on a pull request or release event. Routines run without permission prompts, so scope their connectors carefully. Locally, recent versions start in auto mode, where a second model reviews actions instead of you, and the --dangerously-skip-permissions flag skips nearly all checks; read what it really skips first.
Codex hands work to the cloud from ChatGPT on the web, from a GitHub comment that mentions @codex, or from Slack, and its scheduled tasks repeat work such as a daily drift check. The ChatGPT app also lets the agent keep working while you step away.
Many people run both, one per repo or one per kind of task. Whichever you pick, Briefkin shows every Claude Code and Codex session on one board and pings you on Telegram when one stops for a permission or a question.
Which to choose
- You pay for ChatGPT Plus or Pro already: start with Codex. It costs nothing extra and runs from the CLI, IDE and web.
- You pay for Claude Pro, Max or a Team plan: start with Claude Code.
- You want guardrails that always fire (block edits to
.env, run the linter after each edit, log every command): both can do it; Claude Code has more events and handler types. - You want scheduled or event-driven agents tied to GitHub: Claude Code routines cover schedule, API and GitHub triggers in one place.
- You want to read or patch the agent itself: the Codex CLI is open source.
- You want model choice in one tool: neither; see Cursor vs Claude Code and the full list of Claude Code alternatives.
Questions
Is Codex free?
Codex is included in every ChatGPT plan, Free and Go too, but as of October 2026 those two get it only in the ChatGPT desktop app, with GPT-6 Luna and subject to rollout. The CLI, IDE extension and Codex on the web start with Plus at $20 a month.
Is Claude Code included in the free Claude plan?
No. Claude Code needs a paid Claude plan (Pro, Max, Team or Enterprise) or an Anthropic Console API key. The free plan covers Claude chat only.
Can Claude Code use GPT models, or Codex use Claude?
No. Claude Code runs Anthropic's Claude models and Codex runs OpenAI's models. If you want both families in one tool, look at Cursor, GitHub Copilot or OpenCode.
Can I share one setup between Codex and Claude Code?
Mostly. Both read AGENTS.md, and a skill that sticks to the Agent Skills spec fields works in both: put it in ~/.claude/skills for Claude Code and ~/.agents/skills for Codex. Hooks use different config files, so those you write twice.
Which one is better for long, unattended work?
Both can run in the cloud while you are away. Claude Code adds routines that start on a schedule, an API call or a GitHub event; Codex runs cloud tasks from ChatGPT on the web, GitHub and Slack, and has scheduled tasks in the desktop app.