How many parameters are in GPT-3?
The largest GPT-3 model had 175 billion parameters (175B). The original GPT-3 research paper identifies the 175-billion-parameter model as GPT-3. The paper also describes eight model sizes, ranging from 125 million to 175 billion parameters, so 175B refers to the largest model—not every model in the study. (arxiv.org)
What does the 125-million-to-175-billion range mean?
The GPT-3 study compared models of different sizes to examine how scale affected performance. In that context, 175B is the largest model’s parameter count, while 125 million is the small end of the study’s range. Don’t describe every model in the GPT-3 study as having 175B parameters. (proceedings.neurips.cc)
Does 175B tell you which model to use in a production API?
No. A research paper’s parameter count does not tell you whether a particular model or version is currently offered through an API, or whether it suits your application. Before choosing a production model, check the provider’s current model catalog and documentation for the exact model ID, availability, supported capabilities, and constraints relevant to your use case.
Keep GPT-3 distinct from GPT-3.5 and other model families. For example, OpenAI’s current GPT-3.5 Turbo model page says that GPT-3.5 Turbo was shut down on October 23, 2026; that status is about GPT-3.5 Turbo, not proof of GPT-3’s API status. (developers.openai.com)
Bottom line: GPT-3’s largest model had 175 billion parameters. The original study also evaluated smaller models, and the 175B figure alone does not establish present-day API availability or production suitability.
Sources
Language Models are Few-Shot Learners — arXiv