OpenAI Explained: GPT, ChatGPT and Codex

What GPT actually stands for, how ChatGPT the product differs from the API, what Codex does, and how the subscription ladder compares with per-token billing.

OpenAI is the company most people met first. Its names are also the ones most often mixed up: GPT, ChatGPT, Codex, o-series, Sora. Let us sort them out.

Snapshot, with one caveat

Checked on 30 September 2026. OpenAI's own pricing pages block automated reading, so the plan and model details below come from trade coverage and reference sources and are flagged. Always confirm on the official pages linked at the end.

What the letters mean

GPT stands for Generative Pre-trained Transformer:

Generative: it produces new text (and, in newer versions, images and audio).
Pre-trained: it first learns general patterns from a huge amount of text, then gets specialised.
Transformer: the neural-network architecture introduced in 2017 that all modern large language models use.

If you did our Deep Learning course, you have met all three ideas: pre-training in the revival story, and the Transformer in the history of attention.

GPT is the model, ChatGPT is the product

This is the single most important distinction:

GPT-5.6 (and its predecessors such as GPT-5.5 and GPT-5.4) are models. You reach them through an API.
ChatGPT is the app that wraps those models in a chat interface, with memory, file uploads, search, voice and image tools.
Codex is OpenAI's coding agent, which can run in a terminal, in editors, in a desktop app and in the cloud.

The newest naming: Luna, Terra and Sol

According to reference coverage, the GPT-5.6 family arrived on 9 July 2026 in three variants:

Variant	Role
Sol	the flagship model, the most capable
Terra	a balanced model for everyday work
Luna	the most cost-efficient

The names are Latin words for sun, earth and moon: Sol, Terra, Luna. It plays the same role as Anthropic's Opus, Sonnet and Haiku, and Google's Pro, Flash and Flash-Lite: a big model, a balanced one and a small one.

There is a catch, though: the rank of a name is not fixed between generations. By GPT-6.1, Astra sat above Sol. The full scheme, which name ranks where in each family, and why OpenAI's Astra is not Google's Project Astra, is set out in OpenAI model names: Sol, Luna, Terra and Astra.

What you can do with OpenAI's tools

Chat, write, study and analyse in ChatGPT.
Generate and edit images, and work with voice, through the same app.
Delegate work to an agent: reporting on the launch mentions a ChatGPT Work service in which the flagship acts as an operating agent for longer tasks.
Code with Codex: write features, fix bugs, run tests and open pull requests.
Build products on the API: chat, extraction, search, classification, voice, vision.

Subscription versus API

Again, two separate doors.

ChatGPT subscriptions are flat monthly plans for the app. As reported by pricing trackers, the individual ladder runs from a low-cost Go plan (about $8 a month) through Plus (about $20 a month) to two Pro tiers (about $100 and $200 a month) with much higher usage limits, plus Business and Enterprise seat plans. Check the ChatGPT pricing page for current prices and what each tier unlocks in your country.

The API is billed per token on the developer platform, funded with its own balance. Two sources agree on the key fact: a ChatGPT plan does not include API credits, and API usage is metered separately. Current per-model rates are on the API pricing page.

Practical rule: if you want to use ChatGPT, subscribe. If you want to build something, use the API. If you want both, you pay for both.

Codex: OpenAI's coding agent

Codex is to OpenAI what Claude Code is to Anthropic. The Codex CLI is an open-source (Apache 2.0) agent that runs in your terminal, and it also appears as an IDE extension (VS Code and compatible editors), a desktop app and a cloud agent.

Install, for example on macOS or Linux: curl -fsSL https://chatgpt.com/codex/install.sh | sh, or npm install -g @openai/codex.
Sign in in one of two ways: with your ChatGPT plan (the recommended path, so usage counts against your plan) or with an API key (so usage is billed per token).
What it does: reads your repository, edits files, runs commands and tests, and proposes changes for you to approve.

That two-option sign-in is a pattern you will see across almost every coding agent: use the subscription you already pay for, or bring an API key and pay per token.

Why this matters

OpenAI gives the clearest illustration of model versus product versus agent: GPT-5.6 (model), ChatGPT (product), Codex (agent). Keep that triple in mind and you can read almost any AI headline.