OpenAI Codex
A cloud coding agent that takes a task, works in its own sandbox without supervision, and comes back with a finished pull request for you to review.
What it costs
| Tier | Price | For | Includes |
|---|---|---|---|
| Free | $0/month | Trying Codex without committing to a ChatGPT subscription. | Limited Codex usage bundled into the free ChatGPT tier. |
| Plus | $20/month, billed monthly | One developer using Codex as a regular part of their workflow. | Codex cloud agent access with the usage allowance attached to the ChatGPT Plus plan, plus everyday ChatGPT chat and reasoning models. |
| Pro | $100/month (Pro 5x) or $200/month (Pro 20x) | Developers running Codex heavily enough to hit Plus limits. | 5x or 20x the Codex and ChatGPT usage allowance of Plus. |
| Business | $20/user/month, billed annually ($25/user/month billed monthly) | Teams wanting Codex administered centrally with SSO and billing. | Codex access for the team, central billing, connectors, and admin controls; no training on your code or data. |
| Enterprise | Custom pricing, arranged with sales | Organisations needing compliance controls and credit-based billing at scale. | Credit-based or token-based billing, Codex-only seat types, domain verification, and SCIM. |
Read from OpenAI Codex's own pricing page on 25 Sep 2026. Vendors change prices without notice.
Where it breaks
Codex has no separate subscription — its pricing rides entirely on the ChatGPT plan you already have, and OpenAI retired the old per-message pricing in April 2026 in favour of token-based credits, so how far a plan's allowance actually stretches depends on the size and complexity of the coding tasks you give it. New sign-ups and upgrades to the $200 Pro tier were paused starting 10 September 2026.
First entry in this reference. There is no previous check to compare it against.
Codex is the clearest case here of handing over the work itself: give it a task — fix this bug, add this feature, refactor this module — and it goes to work in its own cloud sandbox, running tests and iterating without you watching each step, then returns with a pull request ready for review. You’re reviewing the diff, not supervising the process that produced it.
There’s no dedicated Codex bill. It’s a feature of your ChatGPT subscription, which means the actual cost of using it seriously is whatever ChatGPT tier you’re already paying for, or would need to upgrade to. Free and Plus give you a taste; anyone running multiple real coding tasks a day will bump into the limits and land on Pro.
What “unwatched” actually means here
Unlike an in-editor assistant that proposes a change and waits for you to accept it line by line, Codex runs a full task end-to-end in isolation from your working copy — cloning the repo, making changes, running whatever tests exist, and only surfacing the result once it has something to show. That’s meaningfully different from pair-programming with an AI; it’s closer to assigning the task to a contractor and reviewing their PR when it lands.
Against the alternatives
Against Windsurf (now shipped as Devin Desktop), the difference is supervision model: Windsurf/Devin Desktop works alongside you inside your editor, turn by turn, while Codex disappears into a sandbox and comes back later. Pick Codex for well-scoped tasks you’re comfortable delegating fully; pick an in-editor agent for anything you want to steer as it happens rather than review only at the end.
The September 2026 pause on new $200 Pro sign-ups is worth checking before you plan around that tier specifically — OpenAI has not said when or whether it reopens.
Get OpenAI Codex
Buying the tool is not the hard part.
The ebook is about what to hand over, in what order, and how you find out when it goes wrong. One payment, no subscription.