GPT
Also known as: Generative Pre-trained Transformer, GPT models, GPT series
GPT stands for Generative Pre-trained Transformer, a name that reflects the model architecture and training approach: pre-train on a large corpus of text to learn language patterns, then fine-tune for specific behaviors. OpenAI introduced GPT-1 in 2018 and GPT-3 in 2020, with GPT-3 marking the moment the research community recognized scale as a path to surprisingly general capabilities. GPT-4 in March 2023 and GPT-4o in 2024 set new standards for multimodal reasoning.
By early 2026, OpenAI had retired the GPT-4 family and the separate o-series reasoning models from the consumer ChatGPT interface, consolidating everything into the GPT-5 line. The GPT-5 architecture integrates reasoning capabilities directly, removing the need for a separate 'thinking' model family. As of May 2026, the current models are GPT-5.5 Instant (the new default for all ChatGPT users), GPT-5.5 Thinking (paid tier, deeper reasoning mode), and GPT-5.5 Pro (highest capability for Pro, Business, and Enterprise plans). GPT-5.3 Codex is the coding-specialized variant powering the Codex agent.
For developers, the GPT family is accessible through the OpenAI Responses API (the current recommended interface, replacing the older Completions and Chat Completions APIs) and through the ChatGPT interface. All current models support a 1 million token context window, multimodal input, tool calling, and structured output. Smaller, cheaper variants like gpt-5.4-mini and gpt-5.4-nano offer lower latency and cost for production workloads that don't need full frontier capability.