GPT (family)

OpenAI's flagship language model family — from GPT-1 (2018) through GPT-4 and the o-series reasoning models. The line that powers ChatGPT.

GPT — Generative Pre-trained Transformer — is OpenAI's flagship LLM family. Each generation has scaled up substantially: GPT-1 (117M parameters, 2018), GPT-2 (1.5B, 2019), GPT-3 (175B, 2020), GPT-3.5 (powering original ChatGPT, 2022), GPT-4 (2023), GPT-4o (multimodal, 2024), GPT-5 (2025). In parallel, the o-series (o1, o3, o4) introduced "reasoning models" with extended internal thinking before answering. It matters because GPT defined the modern era of generative AI. GPT-3's API kicked off the LLM developer ecosystem. GPT-3.5/ChatGPT crossed into mainstream consumer use. GPT-4 raised expectations on what AI could reliably do — passing professional exams, writing production code, reasoning over large documents. Unlike Meta's Llama or Mistral, GPT models are entirely closed source. You can use them only through OpenAI's API or Microsoft Azure OpenAI Service. The exact parameter counts, architecture details, and training data of recent GPT models are not public. The family also includes specialized variants: GPT-4o-mini (cheap fast tier), GPT image / DALL-E (image generation), Whisper (speech-to-text), embedding models, and Codex (now retired but reincarnated as the coding capability inside the main models). Related: OpenAI, ChatGPT, transformer, frontier model, o-series.