GPT means Generative Pre-trained Transformer: a family of large language models made by OpenAI. GPT is the engine; ChatGPT is the chat app built on top of it, the website and phone app you actually type into.
Think of a car and its engine. ChatGPT is the car: the seats, the dashboard, the doors you open. GPT is the engine under the bonnet that makes it go. The same engine can be fitted into other cars, which is why other apps can run on GPT models too.
In a sentence: “The app says it runs on GPT, which means it is built on one of OpenAI’s models rather than on its own.”
What does GPT stand for?
The plain point: each of the three words describes how the model was built.
IBM, the computer company, puts the whole idea in one sentence: “Generative pretrained transformers (GPTs) are a family of large language models (LLMs) based on a transformer deep learning architecture.” It adds that OpenAI introduced the first one, GPT-1, in 2018.1
- Generative: it makes new text, rather than picking an answer from a list. See generative AI meaning.
- Pre-trained: before anyone uses it, it learns the patterns of language from a huge pile of text that nobody has labelled. IBM calls this a form of unsupervised learning.1
- Transformer: the design, or blueprint, the model is built to. It reads a whole passage at once and works out which words matter to each other. See transformer meaning.
GPT vs ChatGPT: what is the difference?
The plain point: GPT is the model; ChatGPT is one product that uses it.
Amazon’s cloud business, AWS, describes GPT models as the technology “powering generative AI applications such as ChatGPT.”2 Other apps reach the same models through an API, a way for software to send a request and get the answer back without a chat window.1
| Word | What it is | Everyday comparison |
|---|---|---|
| GPT | Is: OpenAI’s family of language models. | Like: the engine. |
| ChatGPT | Is: OpenAI’s chat app that runs on those models. | Like: the car built around the engine. |
| GPTs (custom GPTs) | Is: versions of ChatGPT set up for one job. | Like: the same car with a roof rack fitted for one trip. |
The last row trips people up. OpenAI’s help centre says GPTs “are versions of ChatGPT configured for a specific purpose.” The same page says OpenAI plans to retire custom GPTs, with the first business workspaces scheduled for 11 Dec 2026. That is a plan, not confirmed by OpenAI as done as of 5 Oct 2026.3
When someone says “GPT”, ask whether they mean the model, the ChatGPT app, or a custom GPT inside the app. They are three different things with one name.
Who uses the name GPT?
The plain point: mostly OpenAI, which treats the name as its brand.
OpenAI numbers its models with the name, such as GPT-4. Its design guidelines say the “GPT” brand is its property, and that it does not let others put GPT in the names of their apps, products or companies.4 That is OpenAI’s own position as a company.
The letters still turn up in a few older open research models, such as GPT-J from the research group EleutherAI.5 And other chatbots are not GPT models at all: IBM names Anthropic’s Claude and Google’s Gemini as other firms’ own models.1
Where the word comes from
The name grew out of a 2018 OpenAI paper, “Improving Language Understanding by Generative Pre-Training”. Its four OpenAI authors describe “generative pre-training of a language model on a diverse corpus of unlabeled text, followed by discriminative fine-tuning on each specific task”, and say they built it on the Transformer design. The paper’s text does not use the letters “GPT”.6 The short name came afterwards: IBM notes that this first model was “dubbed GPT-1”.1
Often confused with
- LLM: the general kind of model. GPT is one family of LLMs; Claude and Gemini are others. See LLM meaning.
- Transformer: the blueprint. Many models from many companies use it; GPT is just one family built to it. See transformer meaning.
- Generative AI: the wider family of AI that makes text, pictures, sound or video. See generative AI meaning.
Questions people ask
Is ChatGPT the same as GPT?
No. ChatGPT is the app; GPT models are what it runs on, and other apps can use them through an API, as IBM and AWS describe above.
Are Claude and Gemini GPT models?
No. They are large language models made by Anthropic and Google, built on similar ideas but not part of OpenAI’s GPT family. IBM lists them as other firms’ own models.
What is a custom GPT?
A version of ChatGPT given its own instructions, files and tools for one job, in OpenAI’s words “configured for a specific purpose.” OpenAI says it plans to retire them; check its help page for the current position.
Learn more on this site
- The engine type: LLM meaning.
- The blueprint: transformer meaning.
- Picking a chatbot: choosing an AI model.
- Start here: AI explained for complete beginners and the AI glossary.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is the difference between GPT and ChatGPT?
- ChatGPT is the model, and GPT is the app you type into
- GPT is the model, and ChatGPT is the chat app built on it
- They are two names for exactly the same thing
Show the answer
Answer: GPT is the model, and ChatGPT is the chat app built on it. GPT is OpenAI’s family of models, the engine; ChatGPT is the app built on top of it, the car.
2. What does the word “Pre-trained” in GPT describe?
- Each user sets the model up before their first chat
- People check every answer before you see it
- It learns the patterns of language from a huge pile of text before anyone uses it
Show the answer
Answer: It learns the patterns of language from a huge pile of text before anyone uses it. Before anyone uses it, the model learns language patterns from a huge pile of text that nobody has labelled.
3. Are Claude and Gemini GPT models?
- No, they are other companies’ own language models
- Yes, every chatbot is a GPT model
- Only when they run inside ChatGPT
Show the answer
Answer: No, they are other companies’ own language models. Claude and Gemini are large language models made by Anthropic and Google, not part of OpenAI’s GPT family.
Related terms: Chatbot / AI chatbot · Deep learning · Generative AI · Hallucination · LLM (large language model) · Machine learning
- 1IBM, What is GPT (generative pretrained transformer)?, read at source 5 Oct 2026“Generative pretrained transformers (GPTs) are a family of large language models (LLMs) based on a transformer deep learning architecture.” … “AI research firm OpenAI introduced the first GPT model, dubbed GPT-1, in 2018.” ↑a“Generative pretraining is the process of training a large-language model on unlabeled data” … “Generative pretraining is a form of unsupervised learning”. ↑b“These apps connect to GPT through application programming interfaces (APIs), which allow them to pass data back and forth.” ↑c“other firms have released their own generative AI models, including Anthropic’s Claude, Inflection’s Pi and Google’s Gemini, previously known as Bard.” ↑d“AI research firm OpenAI introduced the first GPT model, dubbed GPT-1, in 2018.” ↑e
- 2AWS, What is GPT AI?, read at source 5 Oct 2026“Generative Pre-trained Transformers, commonly known as GPT, are a family of neural network models that uses the transformer architecture and is a key advancement in artificial intelligence (AI) powering generative AI applications such as ChatGPT.” ↑
- 3OpenAI Help Center, GPTs in ChatGPT, read at source 5 Oct 2026“GPTs (also called custom GPTs) are versions of ChatGPT configured for a specific purpose.” … “We’re planning to retire custom GPTs” … “For affected Enterprise workspaces, retirement is planned for Dec 11, 2026.” ↑
- 4OpenAI, Design Guidelines, read at source 5 Oct 2026“the ‘ChatGPT’ and ‘GPT’ brands, and other OpenAI trademarks, are property of OpenAI.” … “we do not permit our GPT brand to be used in app, product, developer or company names”. ↑
- 5EleutherAI, GPT-J 6B model card, Hugging Face, read at source 5 Oct 2026“‘GPT-J’ refers to the class of model, while ‘6B’ represents the number of trainable parameters.” ↑
- 6OpenAI (Radford, Narasimhan, Salimans & Sutskever), Improving Language Understanding by Generative Pre-Training, undated PDF, full text read at source 12 Oct 2026“We demonstrate that large gains on these tasks can be realized by generative pre-training of a language model on a diverse corpus of unlabeled text, followed by discriminative fine-tuning on each specific task.” and “For our model architecture, we use the Transformer”. A text search of the full paper found no use of “GPT”. ↑
The comparisons on this page (the car and its engine, the roof rack) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK