Four OpenAI text-to-speech IDs retire January 6, 2027
OpenAI schedules four exact text-to-speech model IDs for API removal on January 6, 2027 and recommends gpt-realtime-2.1-mini. The replacement model card lists Realtime support, so an existing speech-generation client needs an interface review.
Required action
Inventory the four exact speech IDs. Plan and test the documented Realtime migration before January 6, including connection setup, text-to-audio flow, event handling, voice/output requirements and account access.
Exact IDs and replacements
| Retiring speech ID | Shutdown date |
|---|---|
| tts-1 | 2027-01-06 |
| tts-1-hd | 2027-01-06 |
| gpt-4o-mini-tts-2025-03-20 | 2027-01-06 |
| gpt-4o-mini-tts-2025-12-15 | 2027-01-06 |
One speech notice, four exact IDs
The October 1 deprecation notice sets January 6, 2027 for tts-1, tts-1-hd and two dated gpt-4o-mini-tts snapshots. It names gpt-realtime-2.1-mini as the recommended replacement. Do not extend that table to unlisted speech aliases.
The replacement changes the integration question
The Mini model card lists the Realtime endpoint and WebRTC, WebSocket and SIP connections. The Realtime guide describes sessions and audio events. A migration from a bounded speech-generation request must therefore review session setup, input/output events, audio format, cancellation, voice choices and credential handling. These are preparation steps; Kingy has not run the migration.
Keep adjacent audio deadlines separate
January 6 is not the January 20 legacy audio/realtime notice or the February 26 transcription notice. Check the exact configured ID against the relevant table. An undated alias is not automatically included merely because a similarly named dated snapshot is listed.
Choose an architecture deliberately
OpenAI’s broader Audio and voice guide distinguishes dedicated speech generation, Realtime sessions and GPT-Live conversations. That broader chooser does not replace the Mini recommendation in this exact retirement notice. Validate the required output and supported interface instead of assuming the recommended model preserves every old voice or format.
Evidence limits
Historical first-detection date is unverified. This packet was prepared from retained and newly retrieved official evidence; preparation and retrieval dates are not substituted for first detection.
- No Audio speech endpoint support is established for this recommended Realtime model by its current card.
- No exact voice, file format, latency, cost or output equivalence was tested.
- Scope is the four listed IDs; no family-wide or unlisted-alias shutdown claim.
- January 20 legacy audio and February 26 transcription remain separate events.
- No account access, runtime shutdown or migration run occurred.