Verdict: if you already pay for ChatGPT, Codex is the cheapest way to try a coding agent, because it's included. Start in the desktop app, work in a folder under Git, and keep the training switch off. Pick OpenCode instead if you want your own choice of model.
Other good fits
Tasks you want to start in the cloud and review later, or run from the ChatGPT desktop app instead of a terminal.
+ Pros
Read pros
Codex is OpenAI's coding agent. You give it a goal in plain English, it reads your project, edits files, runs commands and shows you what changed. It comes with your ChatGPT account and runs in several places: the ChatGPT desktop app (switch the dropdown to Codex), ChatGPT on the web with cloud tasks, a terminal tool (Codex CLI) and an editor extension. Plus and higher plans use the new GPT-6 Sol and Luna models.
Pick it when…
You already pay for ChatGPT Plus or Business and want to try a coding agent without another subscription.
You want to start a task in the cloud, close the laptop and review the result later.
You'd rather work in a desktop app than in a terminal.
− Cons
Read cons
Codex can change many files and run commands on your computer. Work only in a folder under Git, and read the plan before you let it act. It works only with OpenAI's models. On personal plans your Codex tasks are used for training unless the switch above is off.
Pick something else when…
You're following the curriculum's coding track or want to choose your own model provider: use OpenCode, taught in T08-L04.
You want an editor where you watch each change as it happens:Cursor has a middle plan at $60.
You pay for Claude rather than ChatGPT:Claude Code does the same job with Anthropic's models.
Avoid if
Your team pays for Claude rather than ChatGPT; Claude Code covers the same ground.
You cannot keep your project under Git; an agent that edits files needs a way back.
Workflow
Choose an available Codex client, sign in with the applicable account, provide a limited task and repository context, review the resulting changes, and run the project's own check. Workspace settings and permissions can constrain available behavior.
About 15–150 GPT-6 Sol or 350–3,000 Luna messages per 5 hours
Pro
from $100
5× (or 20× on the $200 tier) the Plus limits
Business
$20 per user (annual, 2+ users)
Same limits as Plus, admin controls
API key
Pay per use
Billed at OpenAI API prices
In Germany the ChatGPT plans cost €8 (Go) and €23 (Plus). When you hit a limit on Plus or Pro you can buy extra credits or wait for the window to reset. Stay on Plus until you hit the limit several times a week. GPT-5.5 leaves Codex on 14 October 2026, so update any saved settings that name it. The running effort is reviewing every change and keeping the CLI up to date if you use it.
Restrictions
The official help documentation lists ChatGPT account sign-in and specific client setup paths. Enterprise administrators can control starting models, plugins, roles, and some browser capabilities; these controls do not guarantee availability.
First project
Documentation-based suggestion — not run here.
Read first project
Ask Codex to write a small script that renames a folder of files, for example instrument exports or scanned invoices, to a date-first pattern.
Make a new folder with ten copies of made-up files and put it under Git, so every change can be undone.
Check that Settings → Data controls → Improve the model for everyone is off; the switch also covers Codex on personal plans.
Open the ChatGPT desktop app, switch to Codex, pick the folder and ask: "Write a script that renames these files to YYYY-MM-DD_original-name. Show me the plan first and don't run anything until I agree."
Read the plan, let it write the script, then let it do a dry run that only prints the new names.
Run it for real only when the dry run looks right.
Done when all ten files carry the new names, nothing else in the folder changed, and you can undo the whole run with Git.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on OpenAI's Codex documentation and pricing pages (now served from learn.chatgpt.com), the Codex product page, the ChatGPT pricing and data-controls pages and the npm registry (Codex CLI 0.156.1), all read on 2026-09-23. Nothing was run in Codex for this review. Message limits are OpenAI's published ranges; real use varies with task size.
Codex pricing · Sep 23, 2026 — Plans that include Codex, 5-hour message limits per model, extra credits, API-key billing, GPT-5.5 retirement on 2026-10-14.
Refactor repositories, add features, and debug notebooks through terminal coding workflows.
Choose it when an agent should run commands and iterate on test failures.
Its open-source Rust CLI works locally in your own checkout.
OpenAI · paid · cloud
terminal coding
repo editing
test automation
local execution
rust cli
vs code integration
code refactoring
debugging notebooks
write-code
automate
data-analytics
build-app
View sourced imageOfficial Codex code-review screenshot; the publisher's example test result and surrounding presentation are retained.
Codex official product documentation
Owner-requested editorial use of public official interface screenshots, 15 September 2026. Copyright retained by the original publisher; attribution preserved. No open licence or hands-on course execution is asserted. Source asset: https://images.ctfassets.net/kftzwdyauwt9/2yhhXNiYjyEYmc9q5bMHkt/b503f2919724d3be8e6863d9a3ec1403/Codex_Citations_02.png?w=3840&q=90&fm=webp. PNG conversion preserves the complete frame. Converted on 15 October 2026 to a 1600px-wide WebP (quality 80) for faster page loads; the original image remains in the media library.
