Free freeCodeCamp Course: OpenAI Codex Essentials for Agentic Development

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Commands & Code Shown

codex

Purpose: Launches the Codex CLI in the current directory. On first run in a folder it asks whether you trust the contents before proceeding.

When to use: After installing. Installation is a single npm line and, unlike Claude Code, still requires Node.js on your system — Claude Code moved to a standalone binary, Codex has not.

/login

Purpose: Authenticates. Three options: a ChatGPT subscription (the simplest), a device code for one-time-code sign-in, and an API key — first-party or, as the only supported third party, Azure.

When to use: Once per machine. Run it from inside the CLI rather than as a shell subcommand.

/status

Purpose: Shows considerably more than whether you are logged in: usage, which account, what mode, the directory of the currently loaded project, and the current model.

When to use: Whenever a run behaves unexpectedly. Note this differs from Claude Code, which splits login state into a separate auth-status command — here /status is the single diagnostic.

/logout

Purpose: Ends the session and confirms you are logged out.

When to use: On shared machines, or when switching between a subscription and an API-key configuration.

Reproducible steps

1. Install on WSL 2, macOS or Linux — not native Windows

   The recommendation is explicit: install inside WSL rather than on the Windows side, because coding environments are built for Linux and you will otherwise hit unrelated problems. Node.js is a prerequisite.

2. Understand what effort actually controls here

   Codex models support low, medium, high and xhigh reasoning effort. The claim made in the course is that effort determines how large the context window is, separate from the underlying model intelligence — worth verifying against current documentation before you rely on it, since it differs from how effort is described elsewhere.

3. Check the context and output numbers separately

   Both the standard and max Codex models list a 400,000-token context window — so <strong>max is not a bigger context window</strong>, it is described as optimised for long-running tasks. Codex also publishes a max-output-tokens figure separately from context, which Claude's model pages do not, so the headline 400,000 against Claude's 200,000 is not the straight comparison it appears to be.

Gotchas

• This is a free course release from a paid certification platform. Practice exams, cheat sheets and ongoing updates sit behind the paid product, and some content is deliberately held back from the free release. That is disclosed on camera.
• The single most useful line in the course for a Claude Code user: <strong>do not assume your Claude knowledge transfers.</strong> The presenter checks each Codex model page rather than reasoning by analogy, and finds the effort semantics and the context/output split both work differently. That habit is the transferable part.
• Course logistics, if you are considering the certification: exam code EXP-codeex01, 50 multiple-choice questions, 60 minutes, 700 of 1000 points to pass on scaled scoring — so exactly 70% can still fail. Roughly 4-6 hours for an experienced developer, 12-20 if you are new to Codex or to programming.
• The course is explicitly shorter than the equivalent Claude Code course, and the reason given is that there is less surface area in Codex.