Gemini 3.8 Flash TTS is Google's speech model for directing character voices line by line, for audiobooks, games, and voice agents.
If the transcript says "Say cheerfully: Hello", the model may speak that instruction out loud. Sustained delivery goes in speech_metadata.style. A laugh or a pause stays inline. Each turn of a two-speaker scene needs its own speaker name. One request now returns a WAV file. Older Gemini TTS returned raw audio with no header, so a pipeline that adds its own WAV header would wrap the file twice.
Google says Flash TTS is first on Hume AI's Voice Design Benchmark: overall 71.4, accents 60.8. The chart on the same post puts ElevenLabs Voice Design v3 at 70.8 overall and 45.4 on accents, and ahead on voice qualities, 76.6 to 74.6. Those figures are Google's.
The Gemini API pricing page does not list 3.8 TTS. OpenRouter lists Google AI Studio at $0.50 per million input tokens and $9 per million output tokens for Flash, and $6 output for Flash-Lite.
Gemini 3.8 Flash TTS
Ready to try it? Open the official website.