GPT-Audio mini
OpenAI · USA · 2025
The cheap turn-based voice model — four times the context of its realtime sibling, at the same text price.
GPT-Audio mini is OpenAI's low-cost model for audio that does not have to arrive live: it takes speech and text in a single Chat Completions call and answers in speech or text, which suits voice messages, recordings and narration rather than conversation. Its numbers repeat a pattern visible across the family — the turn-based model gets a 128,000-token context and 16,384 tokens of output, four times the window of GPT-Realtime mini, because it never has to hold an open connection. Text costs the same in both: $0.60 and $2.40 per million tokens. The model first appeared as the snapshot dated 6 October 2025 and became the recommended path on 7 May 2026, when gpt-4o-mini-audio-preview was removed. On 20 July 2026 OpenAI put the whole legacy audio family on notice: gpt-audio-mini leaves the API on 20 January 2027, with gpt-audio-1.5 as the replacement — the same successor named for the full-size GPT-Audio, so the cheap tier is not being carried forward under its own name. The October 2025 snapshot was already shut down on 23 July 2026.
▸Videos
No videos yet.
▸Reviews
No reviews yet. Be the first!