All AI agents

OpenAI Codex: the coding agent in ChatGPT, the CLI and the cloud

Codex is OpenAI's coding agent, not the 2021 OpenAI Codex model behind the first GitHub Copilot; Wikipedia keeps the two under separate articles. OpenAI announced Codex's general availability on 6 October 2025, the status recorded above. The CLI within it is open source under Apache-2.0 and written mostly in Rust; the ChatGPT surfaces and Codex cloud are hosted. It also runs as a VS Code extension, in Cursor and Windsurf, with separate Xcode and JetBrains integrations; in the ChatGPT desktop app; and as Codex cloud, which runs tasks in isolated environments you start from the web, GitHub, GitLab in beta, Linear or Slack. TechCrunch dates the CLI's debut to 16 April 2025, the date recorded above; no OpenAI page in our sources dates Codex as a whole. OpenAI moved the docs off developers.openai.com/codex; those links now redirect to learn.chatgpt.com/docs.

#How much does Codex cost?

The pricing page says ChatGPT Work, a separate ChatGPT surface rather than a plan tier, and Codex are included in the Free, Go, Plus, Pro, Business, Edu and Enterprise plans, so there is no separate Codex subscription. Plus is $20 a month and Pro starts at $100. Limits are counted per 5-hour period and vary by model: on GPT-5.6 Luna the page estimates 250 to 2,000 local messages per window on Plus and 1,250 to 10,000 on Pro 5x. Luna is the lightest of the table's three GPT-5.6 models and the most generous by an order of magnitude: siblings Sol and Terra estimate 10 to 100 and 25 to 200 on Plus. The changelog spells Sol GPT-5.6-sol, pricing GPT-5.6 Sol. Local and cloud work share one allowance.

#How do you customise Codex?

A skill is a folder holding a SKILL.md file. Codex reads them from .agents/skills in the repository, ~/.agents/skills, /etc/codex/skills and its own bundled set. The format is shared across vendors: see agent skills. Plugins wrap skills, an MCP server or both behind a plugin.json, scaffolded by the $plugin-creator skill. MCP servers go in an [mcp_servers.<name>] table in config.toml, or via codex mcp add. Hooks live in hooks.json or a [hooks] table and fire on a dozen events from SessionStart to Interrupt. AGENTS.md and AGENTS.override.md are concatenated from the project root down. The wider surface this sits on is ChatGPT.

#How fast does Codex change?

Fast enough that an undated guide misleads. The newest stable CLI was 0.155.1, published on 18 September 2026. The models move too: GPT-5.3-Codex-Spark was deprecated on 14 September 2026, and GPT-5.5 retires from ChatGPT, ChatGPT Work and Codex on all plans on 14 October 2026. Every Codex guide here carries the date it was last checked, as does our Claude Code page.

Key facts

Key facts about OpenAI Codex
MakerOpenAI
TypeCoding agents
StatusGenerally available
Launched
PriceIncluded in every ChatGPT plan, from Free; Plus $20 a month; Pro from $100 a month
AvailabilityCLI on macOS, Linux and Windows; IDE extension for VS Code, Cursor and Windsurf, plus Xcode and JetBrains integrations; ChatGPT desktop and web apps; Codex cloud
Your own skillsYes. SKILL.md folders in .agents/skills, plugin.json plugins, MCP servers in config.toml, hooks.json and AGENTS.md instructions (skills docs)
Websitegithub.com
DocsOpenAI Codex docs
Last verified

Latest news

Frequently asked questions

Is OpenAI Codex free?
OpenAI's pricing page says ChatGPT Work, which is a separate ChatGPT surface rather than a plan tier, and Codex are included in the Free, Go, Plus, Pro, Business, Edu and Enterprise plans. Usage limits reset in 5-hour windows and vary by model and plan, and weekly limits may also apply. Signing in with an API key instead of a ChatGPT account bills at API prices.
Where does Codex look for skills?
In .agents/skills inside the repository, ~/.agents/skills for your user, /etc/codex/skills, and its bundled system skills. In the CLI and IDE you invoke one with the /skills command or a $name mention, and in ChatGPT with @. Codex also picks a skill on its own when a task matches the skill's description.
Does Codex use AGENTS.md or CLAUDE.md?
Codex's docs describe AGENTS.md and AGENTS.override.md, read from ~/.codex and from the project root down to the current directory and capped at 32 KiB by default. Its import tool converts a Claude Code or Cursor setup across: instruction files become AGENTS.md, settings.json becomes config.toml, and slash commands become skills.

Sources

  1. ChatGPT Learn: Pricing
  2. ChatGPT Learn: Build skills
  3. ChatGPT Learn: Build plugins
  4. ChatGPT Learn: MCP
  5. ChatGPT Learn: AGENTS.md
  6. ChatGPT Learn: Import from another agent
  7. ChatGPT Learn: Hooks
  8. ChatGPT Learn: Codex CLI
  9. ChatGPT Learn: Codex IDE extension
  10. ChatGPT Learn: Codex cloud
  11. ChatGPT Learn: Changelog
  12. openai/codex on GitHub
  13. openai/codex releases
  14. OpenAI: Codex is now generally available
  15. TechCrunch: OpenAI debuts Codex CLI
  16. Wikipedia: OpenAI Codex (AI agent)