CodexHowSupport Us

Codex CLI: Defaults, Versions & Commands

Codex CLI — published defaults

Install / update commandcurl -fsSL https://chatgpt.com/codex/install.sh | sh
Observed version0.143.0
Default modelgpt-5.6-sol
Default reasoning effortmedium
Slash commands/init, /status, /permissions, /model, /review
Subcommandscodex, codex resume, codex exec, codex cloud, codex mcp, codex completion
Deprecated flags--full-auto → codex exec --sandbox workspace-write (Still works and prints a warning.)

Everything on this page is a default you can override — but it's worth knowing what actually ships before you start overriding it, since several of these defaults surprise people who assumed differently.

The default model isn't the Codex-named one

The CLI's startup banner names a specific model, and it's the general frontier model, not the model literally named for Codex — that mismatch is this site's most-covered single fact, and it's worth confirming for yourself with a fresh session rather than taking it on faith.

Default reasoning effort

The CLI ships with a medium reasoning-effort setting by default, on the one model where that control is published at all — worth knowing if you're used to a different model's implicit behavior and wondering why effort tuning doesn't appear to do anything on a session running the default model instead.

Install and update

A single command handles both the initial install and subsequent updates on macOS and Linux; Windows installer, npm and Homebrew paths are also offered as alternatives. Whichever path you used to install is generally the right one to use again to update, rather than mixing methods.

Slash commands and subcommands

/init writes a project's AGENTS.md, /status reports session state, /permissions and /model adjust sandbox and model settings mid-session, and /review triggers a review pass. At the shell level, codex, codex resume, codex exec, codex cloud, codex mcp and codex completion cover the major entry points — each with its own flags worth checking via --help on the version you're actually running, since flag surfaces shift release to release faster than any static guide can track reliably.

Checking your actual version

Flag names and defaults documented here are a snapshot, not a permanent contract — codex --version against your installed build is the source of truth the moment anything here seems to disagree with what you're seeing.

Verified 2026-08-09 against https://learn.chatgpt.com/docs/codex/cli.

Checked: https://learn.chatgpt.com/docs/codex/cli · https://developers.openai.com/api/docs/models/gpt-5.6-sol