Gemini TTS migration: request and audio changes to check
A Gemini TTS migration deserves a compatibility check, not just a model-name substitution. Here are the request, audio and budgeting details to review before switching an application.
Google’s Gemini API lifecycle table lists November 17, 2026 for gemini-3.1-flash-tts-preview. The page describes listed dates as the earliest possible retirement dates, with exact shutdown notice provided separately. This is a migration planning date, not evidence that the endpoint has already stopped serving requests.
Choose the replacement before changing the request
The two replacements reached general availability on September 22. Google positions Gemini 3.8 Flash TTS for expressive narration and demanding dialogue, while Flash-Lite TTS targets efficient, higher-volume speech. Both use the same new schema. That compatibility between the two replacements should not be confused with compatibility with the older preview.
Three integration checks that matter
- Separate performance directions from the transcript. Google says text is treated literally. Move sustained style and speaker labels into
speech_metadata; leaving directions in ordinary text can make them part of the spoken output. Keep brief vocal events in supported inline tags. - Identify every speaker turn. Each turn in a multi-speaker request needs a speaker matching the configured voice. Test alternating turns rather than validating only a single sentence.
- Check the audio container. Unary requests now return WAV by default, rather than the older preview’s headerless PCM. Remove a legacy WAV-header wrapper when saving returned WAV bytes. If downstream code requires raw PCM, request that format explicitly.
These changes are documented in Google’s migration guidance. ByteForward has not run an audio benchmark or tested a production migration.
Budget for the date on the rate card
| Standard paid API rate, USD per million tokens | Through December 31, 2026 | From January 1, 2027 |
|---|---|---|
| Flash TTS text input / audio output | $0.50 / $9 | $1 / $18 |
| Flash-Lite TTS text input / audio output | $0.50 / $6 | $1 / $12 |
The pricing page distinguishes text input from audio output. These figures are not per-minute charges or subscription allowances. A forecast should use the application’s actual input and output volumes and the rate period in which it will operate.
Test the complete playback path
The TTS guide specifies text-only input and audio-only output. Single-request multi-speaker generation supports two prebuilt voices; custom voices require separate synthesis by speaker. TTS is distinct from the bidirectional Live API.
Use a short representative script with a name, a number, a pause and alternating speakers. Check pronunciation, file decoding, concatenation and playback in the application that users will actually hear. Keep the old configuration recorded for comparison, but do not assume a deprecated endpoint will remain a dependable long-term fallback.
AI-generated editorial illustration by ByteForward with the official Gemini wordmark from the Gemini website. Gemini is a trademark of Google LLC. ByteForward is not endorsed by or affiliated with Google.
