GPT.

Generative Pre-trained Transformer — the model architecture and family name behind OpenAI's most famous models, from GPT-2 to GPT-5.

GPT stands for Generative Pre-trained Transformer. 'Generative' means the model produces new text. 'Pre-trained' refers to the initial training on large text corpora. 'Transformer' names the underlying neural network architecture introduced in 2017. OpenAI released GPT-1 in 2018, GPT-2 in 2019, GPT-3 in 2020, and GPT-4 in 2023.

The GPT series demonstrated that scaling a simple architecture — predict the next token — on enough data produces remarkably capable models. GPT-3's 175 billion parameters was a milestone that showed few-shot learning emerging naturally from scale, without any task-specific training beyond in-context examples.

Today 'GPT' has become a generic term that many non-technical users apply to any AI chatbot. Within the field, it specifically refers to OpenAI's decoder-only transformer models. The GPT-5 generation, released in 2025, represents OpenAI's most capable models to date, with strong reasoning, long context, and multimodal capabilities.