OpenAI released GPT-3 in 2020 with 175 billion parameters.
GPT-3 is a generative pre-trained transformer language model. It was described in OpenAI's paper “Language Models are Few-Shot Learners,” published in 2020. The model generated text by predicting likely continuations from preceding text, and its very large scale allowed it to perform many tasks from prompts and examples without task-specific retraining.
The 175-billion figure refers to the model's learned parameters, not the number of words it knows or the number of training documents. Parameters are numerical values adjusted during training, and their count is only one measure of a model's size and capability.
GPT-3 is sometimes confused with ChatGPT, which is a conversational application and model family built later. It is also distinct from GPT-2, whose largest released version had 1.5 billion parameters, and GPT-4, released in 2023 with a different public description of its size. GPT-3's scale helped accelerate public interest in large language models.