GPT-3 Davinci is the largest of the four original GPT-3 model sizes OpenAI made available through its API when it launched in June 2020, alongside the smaller Ada, Babbage, and Curie tiers. At 175 billion parameters it was, at the time, the biggest language model publicly accessible through an API, with a context window of 2,049 tokens. OpenAI never released its weights; access was API-only and metered by usage.
Davinci is what powered the wave of early GPT-3 demos, from auto-generated code snippets to essay writing, that convinced much of the tech industry that large language models had crossed a usability threshold. Later variants such as text-davinci-002 and text-davinci-003 added instruction tuning on top of the same base. OpenAI retired the model on January 4, 2024, as part of a broad deprecation of its original GPT-3 lineup, directing remaining users to gpt-3.5-turbo-instruct.