GPT Glossary
GPT stands for Generative Pre-trained Transformer — OpenAI’s family of large language models. The series began with the original GPT (June 2018, 117M parameters, a research artefact), proceeded through GPT-2 (February 2019, controversial open-release decisions), GPT-3 (June 2020, 175B parameters, the first model that meaningfully demonstrated few-shot learning), GPT-3.5 (the model underneath the original ChatGPT), GPT-4 (March 2023), GPT-4o and GPT-4-Turbo (2024), and GPT-5 (2025) with ongoing minor revisions.
The transformer architecture GPT’s name refers to was introduced in Vaswani et al.’s 2017 paper Attention Is All You Need, originating at Google. OpenAI’s GPT line was the first sustained demonstration that the architecture, scaled aggressively with the right training data and compute, produced broadly useful general-purpose text generation. The success of GPT-3 in particular triggered the post-2020 LLM era and the competitive race that produced Claude, Gemini, Llama, and the rest.
For this Dictionary, the GPT family matters primarily as the market comparator — the model series the rest of the industry’s products are positioned against. The operator’s daily working model is Claude (specifically Opus), but GPT-class models are the structural reference that informs nearly every comparison readers will draw.