deniz.in

Markets

Weather

Loading weather

· via dev.to (home feed)

Google flags November 17 as earliest shutdown for five Gemini audio preview models

Google's deprecation page lists November 17, 2026 as the earliest shutdown date for five Gemini audio preview models, meaning TTS and Live API calls could start failing from that day onward.

Google flags November 17 as earliest shutdown for five Gemini audio preview models

Google sets November 17 as earliest cutoff for Gemini audio previews

Google's deprecation page gives five Gemini audio preview model IDs an earliest shutdown date of November 17, 2026, according to a post on dev.to by Versyka. Three of the affected models serve text-to-speech workloads and two handle Live, real-time voice sessions. Applications still calling the old IDs could begin failing as soon as that date passes.

The five model IDs and where they move

Versyka lists the affected identifiers together with the replacements Google points to:

Model ID Suggested replacement
gemini-3.1-flash-tts-preview gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-2.5-flash-preview-tts gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-2.5-pro-preview-tts gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-3.1-flash-live-preview gemini-3.8-live
gemini-2.5-flash-native-audio-preview-12-2025 gemini-3.8-live

"Earliest" is the operative word

The November 17 date is not a fixed retirement deadline. As Versyka explains, it marks the first day the models could be switched off, and calls may stop on any day from that point onward without a further announcement. Preview models also tend to receive shorter deprecation notice than stable ones, because Google never committed to keeping them available. In practice that leaves no safe window after the date passes, which makes the date already on the calendar the only reliable deadline.

What to check and change

Versyka recommends a few concrete steps before the cutoff arrives:

  • Search the codebase and configuration for the affected strings, for example with grep -rnE "preview-tts|native-audio-preview|flash-live-preview" . and remember to check environment variables and anywhere a model name is stored as data rather than in code.
  • Point TTS traffic at gemini-3.8-flash-tts, or at gemini-3.8-flash-lite-tts where cost matters more than output quality, and move Live sessions to gemini-3.8-live.
  • Actually listen to the generated audio. Voices, pacing and pronunciation shift between model generations, and a test that only asserts audio came back will not catch a voice users no longer recognise.
  • Run at least one complete Live session end to end, since session settings that worked against the preview models may not carry over unchanged.

Why it matters

Voice features are frequently built as afterthoughts: a read-aloud button, a voice mode added for a demo, a live-agent prototype that quietly became a shipped feature. Because they are rarely the core of a product, nobody tends to own the migration until the first bug report arrives. A shutdown that can begin without further notice turns that neglect directly into an outage.

The episode is also a reminder of the general bargain that preview model IDs represent: earlier access in exchange for no longevity guarantee. Teams running any preview model in production should treat provider deprecation pages as a monitoring input rather than something to read after the errors start. Versyka, which also builds tooling that tracks provider shutdown dates and flags affected model IDs in code, frames the goal as turning the swap into a reviewed pull request instead of an incident.

  • #gemini
  • #google-gemini
  • #text-to-speech
  • #api
  • #deprecation

Related posts