GPT-4o mini

GPT-4o mini arrived on 18 July 2024 as the small member of the omni family: text and image in, text out, a 128,000-token context window and a 16,384-token output limit, with a knowledge cutoff of 1 October 2023. It is the cheapest OpenAI model of its generation at $0.15 per million input tokens and $0.60 per million output, with cached input at half the input rate. Its role in practice was distillation and fine-tuning. OpenAI positioned it explicitly as a target for outputs from a larger GPT-4o model, so that a team could train the small model on the big one's answers and get comparable behaviour at a fraction of the cost and latency. It remains one of the few OpenAI models where fine-tuning is supported directly, and from 28 October 2024 it also replaced the legacy babbage-002 and davinci-002 as the recommended base for new fine-tuning runs. Its durability is the surprise. Of the OpenAI models released in 2024, GPT-4o mini is the one that has neither been retired nor been given a shutdown date: GPT-4 Turbo, o1 and the GPT-4o audio and realtime previews all have theirs. Its search-preview variant was shut down on 23 July 2026, but the base model is still listed in the API.

▸Videos

No videos yet.

▸Reviews