GPT Glossary
GPT stands for Generative Pre-trained Transformer, the name of OpenAI’s principal model lineage. The original 2018 research model showed that a transformer pretrained on a large text corpus could be adapted to many language tasks. GPT-2 extended the approach in 2019; GPT-3 demonstrated strong in-context and few-shot performance at much larger scale in 2020; GPT-3.5 powered the first public version of ChatGPT; GPT-4 followed in 2023; and the GPT-5 series began in 2025. Individual product and API names change more quickly than the family definition.
The transformer in the name refers to the architecture introduced by Vaswani and colleagues in the 2017 paper Attention Is All You Need. GPT did not invent the transformer. Its historical importance lies in making generative pretraining plus scale a durable research and product programme. That programme helped establish the modern large-language-model market in which Claude, Gemini, Llama, and other families now compete.
For this Dictionary, GPT matters primarily as the market comparator: the model lineage against which many other products are positioned. It is also useful to distinguish GPT, a model family, from ChatGPT, a consumer and workplace product that can route among models and tools.
Sources
- Radford et al., “Improving Language Understanding by Generative Pre-Training”, 2018.
- Brown et al., “Language Models are Few-Shot Learners”, 2020.
- OpenAI, “Introducing GPT-5”, August 2025.