GPT-4o mini Audio, available since 17 December 2024, is the budget member of OpenAI's turn-based voice family. It does what GPT-4o Audio does — takes a recording, thinks, answers in speech through an ordinary chat completion — but charges $10 per million audio tokens in and $20 out, a quarter of the flagship rate, and $0.15 / $0.60 for text. The savings do not come out of memory: it keeps the full 128,000-token context window and the same 16,384-token maximum answer, so the difference the developer is buying is the quality of the underlying model, not its capacity. Knowledge ends on 1 October 2023 and the model stayed in preview throughout. It was swept up in the 20 July 2026 retirement announcement covering the legacy voice line and stops answering on 20 January 2027; OpenAI points users to gpt-audio-1.5.