A Generative Pre-trained Transformer (GPT) is a type of AI language model made by OpenAI that predicts text one piece at a time and, by doing that very well, can write, summarise, translate and answer questions. The name spells out how it works: it is generative because it produces text, pre-trained because it learned from a massive body of writing before you ever used it, and a transformer because of the network architecture inside.
The simplest way to picture it is an extremely well-read autocomplete. Your phone guesses the next word in a text message; GPT does the same trick, but trained on so much material and at such scale that the “next word” can extend into coherent paragraphs, working code and reasoned arguments. It is the most famous example of a large language model.
GPT is not one model but a line of them, each generation bigger and more capable than the last. The early versions could barely hold a thread across a paragraph. Later ones handle long documents, follow detailed instructions, and call external tools to look things up or run calculations they cannot do reliably on their own.
People often use GPT and ChatGPT interchangeably, but they are different layers. GPT is the raw engine. The chat app is the car built around it. Developers can skip the app entirely and call GPT directly to power their own software, paying per token of text it reads and writes rather than a flat subscription.
At TopDevs we build on GPT models through their API when a project needs strong text understanding, and we choose between GPT and rival models per case rather than assuming one is always best.
