OpenAI davinci-002

davinci-002 is the larger of the two base models OpenAI kept after clearing out the GPT-3 shelf, standing in for both curie and davinci when those were shut down on 4 January 2024. Like its smaller sibling it completes text and nothing else: no instruction following, no chat, no tools, and only the legacy Completions API. The price is where its history shows. The original davinci cost 20 dollars per million tokens; davinci-002 costs 2 — a tenfold cut for a model sold as its direct replacement, and a reminder of how far inference costs fell between 2020 and 2023. Input and output are billed at the same rate, a convention that disappeared from OpenAI's price list everywhere else. Output runs to 16,384 tokens and the knowledge cutoff is September 2021. OpenAI stopped accepting new fine-tuning runs on it on 28 October 2024, deprecated the model on 26 September 2025 and will shut it down on 28 September 2026, pointing users to gpt-5.6-terra. When it goes, the last model in the API descended directly from the GPT-3 base line goes with it — six years after that line defined what a large language model was.

▸News

OpenAI told nine models to move here — and three of those destinations switch off on 28 September

8/20/2026 · OpenAI — strona wycofań modeli (deprecations)

▸Videos

No videos yet.

▸Reviews