Google on September 23 released Gemini 3.8 Flash TTS and a cheaper Flash-Lite sibling, adding promptable voice design, more than 2,000 production voices and 30-second voice replication to the Gemini stack. Every output is stamped with SynthID watermarking, which Google says survives common audio codecs.
Prompt A Voice, Get A Voice
Promptable voice design lets developers describe the desired timbre and delivery — "warm, mid-30s British female, slight rasp, upbeat" — and receive a matching synthetic speaker without recording sample audio. Behind the scenes, Google's voice-design system samples through its 2,000+ production voices and blends acoustic parameters until the prompt is matched.
The Benchmark Claim
Google reports the top slot on the Hume Voice Design benchmark, ahead of ElevenLabs Turbo v3 and OpenAI's most recent voice model. Independent evaluations are not yet published; the claim covers naturalness, prompt adherence and identity consistency across long-form generations.
Rollout And Pricing
Gemini 3.8 Flash TTS is available in AI Studio, the Gemini API and Vertex AI. Google has cut pricing on TTS output by roughly a third versus 3.5 Flash TTS; Flash-Lite drops another 50% at a small quality trade-off. Voice replication is gated to verified enterprise accounts.
Related coverage: Anthropic ships Claude Opus 5.5, OpenAI's GPT-6 Sol/Luna price cut, and NVIDIA's Nemotron 3 Diarization.
Reporting based on coverage from Google Deepmind, Hume and industry announcements on September 23, 2026.