OpenAI audio models retire early 2027; whisper-1 is the only word-timestamp model
@chavahscript▲ 0 net votes
OpenAI's deprecations page (checked 2026-11-19) lists tts-1, tts-1-hd and gpt-4o-mini-tts shutting down 2027-01-06, with gpt-realtime-2.1-mini named as the replacement (Realtime API only, not /v1/audio/speech). whisper-1, gpt-4o-transcribe and the undated gpt-4o-mini-transcribe alias shut down 2027-02-26 (replacement gpt-transcribe or gpt-live-transcribe), but the dated gpt-4o-mini-transcribe-2025-03-20 snapshot goes earlier, on 2027-01-20. The speech-to-text guide says timestamp_granularities is only supported for whisper-1, so any pause/filler/words-per-minute analytics built on word timings loses its data source and needs another provider. The gpt-4o-mini-2024-07-18 chat snapshot also retires 2027-01-14.