Model
GPT
OpenAI's Generative Pre-trained Transformer model family, from the 2018 original to GPT-6 Astra, which powers ChatGPT and the OpenAI API.
GPT, for Generative Pre-trained Transformer, is the name OpenAI has used for its main line of language models since 2018. The approach, described in a June 2018 OpenAI post, pairs a transformer network with unsupervised pre-training on large amounts of text, followed by fine-tuning for specific tasks. Successive generations scaled this recipe up and added instruction following, chat, tool use, vision and reasoning.
The family is the foundation of ChatGPT, the consumer assistant launched in late 2022, and of the OpenAI API used by developers and enterprises. It also powers Codex, OpenAI’s coding agent.
The current generation is GPT-6. On 3 September 2026 OpenAI began rolling out GPT-6 Astra, which it describes as a generational step toward artificial general intelligence, first to business customers in its Daybreak early-access program and then to paid ChatGPT tiers with added cybersecurity guardrails. OpenAI says Astra is aimed at computer use, browsing, software engineering, cybersecurity, science and professional work. On 7 October 2026 the company published an announcement titled “GPT-6 and Intelligent UI for everyone,” and its model listings include further GPT-6 variants, among them GPT-6.1 Sol and GPT-6 Luna, alongside GPT-5.5 and GPT-5.6.
GPT models are offered as closed weights through ChatGPT, the OpenAI API and partners such as Microsoft and AWS. In ChatGPT they are paired with voice conversation, image generation and web search. OpenAI has not described any GPT model as a Super Intelligence.
For current models and pricing see openai.com.