Is Codex Free? What ChatGPT Free Gets, What Needs a Paid Plan, and the $0 Local Route We Tested
Codex is included in ChatGPT Free and Go, but not Codex Cloud or image generation, and OpenAI publishes no Free usage numbers. We tested Codex CLI 0.167.0 without a login, with a made-up API key and against local Ollama.

Is Codex free? Yes to start with, but only a thin slice of it. OpenAI’s help center says Codex is included in every ChatGPT plan, Free and Go included. The Codex pricing page lists the Free plan at $0 with GPT-6 Luna at standard speed in the desktop app, “subject to rollout”. Codex Cloud, image generation and the Plus-level usage estimates are not part of it. This guide separates what is free, what is not, and what we tested on Codex CLI 0.167.0 on November 10, 2026.
The short answer
- ChatGPT Free and Go: Codex is included. The pricing page lists one thing for both: GPT-6 Luna at standard speed in the desktop app, subject to rollout.
- Codex Cloud: the help center says it is not included with Free or Go.
- Image generation: the pricing page says it isn’t available on the Free plan.
- Usage numbers: the pricing page gives local-message estimates for Plus and Standard Business only. It publishes no numbers for Free, so read your own limit in the product.
- The Codex CLI itself: free to install and open source under Apache-2.0. It still needs a sign-in, an API key or a local model to do anything.
- API key: you pay API prices. We did not find a free tier on that route.
What each plan lists

Two sources disagree a little on where Free can run Codex. The pricing page lists Free and Go with the desktop app only, and puts web, CLI, IDE extension and iOS under Plus. The help center’s getting-started list names the desktop app, the CLI, the IDE extension and Codex web as clients you can sign in with, without separating plans. We could not settle it, because the box has no ChatGPT account. If you are on Free, check inside the app you want to use.
OpenAI also says Codex usage limits vary by plan and that you can see yours in the usage dashboard, or with /status in a CLI session. For Plus and Business the pricing page gives estimates per 5 hours, for example 350 to 3,000 local messages on GPT-6 Luna. Treat them as a ceiling for Plus, not as a promise for Free. The help center adds that desktop users may be asked to add or verify a phone number.
The CLI is free, the access is not
The open-source part is real. The openai/codex repository is under the Apache-2.0 license and npm install -g @openai/codex installs it. What it cannot do is run a model for you. We installed 0.167.0 on a clean box and tried it three ways.

- No sign-in:
codex login statusprinted “Not logged in” and exited with code 1.codex exectried a WebSocket five times, fell back to HTTPS and ended with401 Unauthorized: Missing bearer or basic authentication. - API key route: we saved a made-up key with
codex login --with-api-key. Codex accepted it (“Authentication is checked when you send a request”) and the first request came back401 Incorrect API key provided. A real key would be billed at API prices, as the Codex pricing page says for the API Key option. - ChatGPT sign-in:
codex loginopens a browser flow. We did not run it, so we cannot say what a Free account sees in the CLI.
The $0 route: Codex CLI with a local model
The one route that costs nothing and needs no OpenAI account is pointing the CLI at a local Ollama server. We tried it: Ollama 0.40.16 on CPU, model llama3.2:1b, and -c 'model_provider="ollama"'. Codex found Ollama on its default port 11434 and sent the request to its /v1/responses endpoint, which answered 200 after 21 seconds.
The result was not useful. Codex reported 3,890 tokens used for one small question, and the 1B model replied with a broken tool-call JSON instead of an answer. So the plumbing works, but a model that small cannot drive an agent. In our run, --oss --local-provider ollama failed with a provider mismatch error on this build, while setting model_provider worked.
Two practical limits follow. One small question already cost 3,890 tokens, so a small context window fills fast; see Ollama context length. And you need a model that handles tool calls well, which is a hardware question; our local model guide covers sizes. We only tested a 1B model, so we make no claim about how larger local models behave.

If Free runs out
Plus is $20 a month on the pricing page and lists Codex on the web, in the CLI, in the IDE extension and on iOS, plus credits to extend usage. Go is $8 and, on the pricing page, has the same Codex line as Free. Pro starts at $100. To see where Codex sits in your whole AI bill, use the AI subscription cost calculator. For the differences between the ChatGPT tiers, see ChatGPT Go vs Plus vs Pro, and for the limits themselves how to check Codex usage limits.
If you are choosing between coding agents rather than plans, our Claude Code vs Codex comparison and the CLI vs app guide are the next stop; install steps are in Codex CLI install.
What we tested and what comes from the docs
- Tested on November 10, 2026: Codex CLI 0.167.0 on Linux with a clean config each time. Not logged in, a made-up API key, and a local Ollama 0.40.16 server on CPU with
llama3.2:1b. The outputs in the image are copied from those runs. - From the docs, not tested: everything about the Free and Go plans (models, desktop app, Cloud, image generation, phone verification) and all usage numbers. We have no ChatGPT account on the box.
- Not tested: larger local models, GPU speed, and Codex with a real API key.
- Sources: OpenAI’s Codex pricing, the Help Center article Using Codex with your ChatGPT plan, the authentication docs, the CLI docs and the openai/codex repository. Plans and models change, so recheck before you rely on a detail.
FAQ
Is Codex free?
Partly. OpenAI says Codex is included in ChatGPT Free and Go, with limits. The pricing page lists GPT-6 Luna at standard speed in the desktop app for both. Cloud tasks and image generation are not part of it.
Is the Codex CLI free?
The software is free and open source under Apache-2.0. Running it needs a ChatGPT sign-in, an API key billed at API prices, or a local model.
How many Codex messages does the Free plan get?
OpenAI does not publish a number on the pages we read. The estimates on the pricing page are for Plus and Standard Business. Your own limit shows in the usage dashboard.
Can I use Codex without an OpenAI account?
Only with a local model. With no sign-in, the CLI gets a 401 from the OpenAI API. With Ollama it ran in our test, though a 1B model gave no usable answer.
Is Codex Cloud free?
No. The help center says Codex Cloud is not included with Free or Go.