Codex · Series · Part 4 of 8

Codex models, the API and official links

Parts 1 to 3 were about using Codex well. This part is the reference shelf at the end of the beginner level (Parts 5 to 8 go further): how Codex relates to the OpenAI API and the two ways to sign in, a first API call, the models you can pick in Codex as the docs list them on 18 October 2026 (including the ones that have just been retired), how to choose a model and a reasoning effort, how plans and pricing fit together, and the official pages I rely on, all in one place. Models and plans change often, so every table here is a dated snapshot and the linked docs are the source of truth.

1. Codex or the OpenAI API?

The same models can be reached through different doors. Which one you want depends on what you are trying to do. The docs describe Codex as working in the ChatGPT desktop app, the Codex CLI and the IDE extension, plus Codex cloud for tasks that run on the web. The OpenAI API is a separate door for building your own software.

Sign in with ChatGPT or with an API key

The authentication docs say Codex supports two ways to sign in when using OpenAI models, and that the ChatGPT desktop app, the Codex CLI and the IDE extension support both for local work:

In the CLI, the commands from the docs are:

# ChatGPT sign-in: opens a browser window
codex login

# API key sign-in: pipe the key in from an environment variable
printenv OPENAI_API_KEY | codex login --with-api-key

# Show the active sign-in method (exits 0 when you are logged in)
codex login status

If you work on a remote or headless machine where the browser flow does not work, the docs describe a device-code option (codex login --device-auth, marked beta). The docs also recommend API keys for programmatic CLI workflows such as CI jobs, and warn you not to expose Codex execution in untrusted or public environments. Part 7 comes back to automation.

2. A first API call

This is the same idea as the context window from Part 3: the API does not remember anything between calls, so your code sends what the model needs every time. The docs call the main endpoint the Responses API. I checked the steps below against the live developer quickstart on 18 October 2026.

1. Create an API key in the OpenAI dashboard (the quickstart has a button for it) and store it somewhere safe. Treat it like a password.
2. Put the key in an environment variable so it never appears in your code. On macOS or Linux:

   export OPENAI_API_KEY="your_api_key_here"

   On Windows the quickstart uses setx OPENAI_API_KEY "your_api_key_here". Each official SDK reads this variable by itself.
3. Install the SDK: pip install openai for Python (or npm install openai for JavaScript).
4. Make a call. Create example.py with the code from the quickstart, then run python example.py:

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-6-astra",
    input="Write a one-sentence bedtime story about a unicorn.",
)

print(response.output_text)

The JavaScript version in the same quickstart does the same with client.responses.create(…) and prints response.output_text. The quickstart also shows examples for .NET, Java, Go and Ruby, and plain HTTP examples that send a request to https://api.openai.com/v1/responses with an Authorization: Bearer header. For the exact request and response shapes, use the Responses API reference rather than my summary.

3. Models in Codex (as of 18 October 2026)

This is how the Codex docs list the models on 18 October 2026. OpenAI changes the lineup often, so treat it as a snapshot and check the Models page, the API model catalog and the dated What’s new page for the live status. The docs also stress that availability depends on the rollout, your sign-in method, your client, your plan and your workspace settings.

Models to pick from

The docs also say GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna remain available during the rollout. In ChatGPT, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna are available in Work and Codex, not in Chat. The GPT-6.1 Sol launch rollout covers Plus, Pro, Business, Enterprise and Edu in Codex in the desktop app and CLI; for Enterprise and Edu it is off by default until an administrator enables it, and Free and Go were not included at launch. Selecting a new model does not change workspace permissions or give you access to it.

Retired and deprecated in Codex

The “What’s new” and Models pages carry dated retirement notices. If an old script or config still names one of these, update it.

The pattern in the docs is the same each time: retirement applies to the ChatGPT sign-in path, and you should replace the model name in workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks and scripts. With an API key, model availability follows your API organisation and project instead, and the docs note that GPT-5.5 remains available through the API.

4. Switching models and reasoning effort

If you do not choose, Codex uses a recommended model. The docs describe these ways to choose one yourself:

model = "gpt-6.1-sol"
model_reasoning_effort = "medium"

The desktop app, the CLI and the IDE extension share the same config.toml. The docs list the model_reasoning_effort values low, medium, high, xhigh, max and ultra, and add that the levels available depend on the model and client. Command-line flags and --config overrides beat the file for a single run. If you do not have access to the model in an example, use one you can see in your picker.

5. Choosing a model and an effort

The docs give a clear starting point: Luna is the most cost-efficient model, Astra the most powerful. If you do not need to think about usage or time, you can default to Astra. Otherwise, choose by the job:

Reasoning effort

Reasoning effort is how much thinking the model does before answering. Higher effort can improve results on complex tasks, but it takes longer and uses more tokens (Part 3). The docs say to start with the default and raise it when the task needs deeper planning or analysis.

Two honest caveats from the docs. Reasoning efforts do not map exactly between model generations, so try a familiar task at a lower setting and adjust. And the guidance is a starting point: experiment with the same input, compare, and keep the lightest setting that meets your quality bar. Ask yourself how often the task runs, how fast you need it, how the output will be used and how much quality matters. The docs also describe Standard, Fast and Ultrafast speed modes where supported, which use your allowance at different rates (see the Speed page).

6. Plans and pricing, at a high level

Prices and limits change more often than anything else in this post, so I only describe the shape here. The Codex pricing page is the source of truth. It says ChatGPT Work and Codex share usage, and that API token prices are separate from subscription usage, so you should not use API prices to estimate how much a plan includes.

Usage depends on the model, the size of the task, the context, tool use and caching, so the same-looking tasks can cost different amounts. Smaller, cheaper models such as Luna go further on a plan than Astra. The pricing page also has a usage estimate table, credit rates and answers to questions like “where can I see my current usage limits”. Check it, and your own usage dashboard, rather than trusting any number copied into a blog post (including mine).

7. Official links

Codex changes quickly. This series was checked against the official documentation on 18 October 2026; always confirm details on the official pages. Every link below returned HTTP 200 when I checked it. If a link breaks, the docs are indexed at learn.chatgpt.com/docs and the API docs at developers.openai.com.

Codex: getting started and sign-in

Models, pricing and what changed

OpenAI API

Open source, community and support

8. Beginner-level wrap-up and glossary

1. Part 1: set up Codex safely, learn the modes and sign in.
2. Part 2: give Codex lasting context with AGENTS.md.
3. Part 3: spend fewer tokens, hand work off between sessions and onboard faster.
4. Part 4: know your sign-in options, the model list and the official links.
5. Next (advanced): Part 5 covers working with Codex day to day.

Models are retired, plans move and the docs change often. Writing things down, and checking the official pages when something looks off, are the two habits that carry through all four parts.