Voice API deadlines, October 2026 to February 2027
A calendar of voice and transcription API changes from vendor pages: Gemini TTS price rise, OpenAI transcription shutdown, and the xAI voice alias move.
Two dates matter for anyone with a voice integration this winter. Gemini's 3.8 Flash TTS prices double on 1 January 2027, and OpenAI's whisper-1 and three gpt-4o transcription models shut down on 26 February 2027. xAI already moved its grok-voice-latest alias on 5 August 2026. Check the calendar before you ship another release.
The calendar
| Date | Vendor | Change |
|---|---|---|
| 5 Aug 2026 (past) | xAI | grok-voice-latest moved to Think Fast 2.0; users who had not pinned were upgraded |
| 26 Aug 2026 | OpenAI | Deprecation notice: whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize |
| 31 Dec 2026 | Gemini 3.8 Flash TTS at $0.50 text input and $9.00 audio output per 1M tokens ends | |
| 1 Jan 2027 | Flash TTS becomes $1.00 and $18.00; Flash-Lite TTS $1.00 and $12.00 (from $0.50 and $6.00) | |
| 26 Feb 2027 | OpenAI | Shutdown of the four deprecated transcription models; move to gpt-live-transcribe or gpt-transcribe |
What to do on each
- Pin explicit model versions today. Aliases that say latest move without warning.
- Budget the January TTS price step now if you use Gemini TTS at volume.
- Run your transcription test set on the replacement model before February, not in the last week.
The same discipline on Sume
Sume's TTS router has sonic-latest next to explicit ids such as sonic-3.6; pin an explicit id if you do not want output to change under you. Music 1.0 is retiring gradually and resolves through the music router, so new work should call the router. The API reference lists the live model ids, and the catalog endpoint shows what is available now.
Limits
Vendors reschedule. These rows are what each page said on 1 October, and they will age. Prices are quoted from the vendor pages, not verified by a test call.
Sources
More in Developers
- Wait for an avatar video job with the SDK: waitForJob and its timeout
Avatar jobs need waitForJob, not waitForRun. Defaults, the 20-minute timeout, and why a SumeJobTimeoutError neither cancels nor refunds the render.
- waitForRun and 429 or 503 on a status read: it keeps polling
A 429 or 5xx on a status read does not fail the run. waitForRun tolerates 6 consecutive transient read failures and keeps waiting. Options and what to log.
- Wan 2.2 TI2V-5B in Diffusers: 121 frames, 24 fps, 24GB GPU
A working Diffusers example for Wan2.2-TI2V-5B: 704x1280 portrait frames, 121 frames at 24 fps, 24GB VRAM minimum, and when to use a hosted route instead.
- How many Wan 3.0 clips run at once on Sume: 1, 4, 8 or 20
Wan 3.0 jobs run under plan-only concurrency: Free 1, Pro 4, Startup 8, Scale 20. Extra jobs queue. What that means for a 30-second render batch.
Written by Sume