GPT-1 is a language model with 117 million parameters, GPT-2 has 1.5 billion, GPT-3 has 175 billion, and the performance of the language model is improving as the number of parameters increases.