Google launches Gemini 3.8 Flash TTS and Flash-Lite TTS for custom AI voices

Google has released two new text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, aimed at developers and companies that need custom, expressive synthetic voices. Announced on September 23, 2026, both are available today through the Gemini API and Google AI Studio.

Two models, two jobs

What the models can do

According to Google's announcement, the new models add:

• Generative voice design: describe a voice in plain language and get a custom one, across 100+ languages and dialects.
• A large voice library: 2,000+ production-ready voices, including regional varieties such as Mexican Spanish, Quebec French, Scots English and Brazilian Portuguese.
• Voice replication: recreate a voice from a 30-second audio sample. This requires a recorded verbal consent check.
• Line-by-line direction: control acting cues, pacing and emotional tone for each line.
• Long-form audio: Google says quality holds across hours of continuous audio.
• Two-speaker scenes: native support for directing conversations between two voices.
• Vocal bursts and backchanneling: laughs, sighs, gasps and short interjections.

Google also lists voice remixing, which fine-tunes timbre, pitch, pace and accent, as coming soon.

Google's benchmark claims

Google says Gemini 3.8 Flash TTS ranks first on Hume AI's Voice Design Benchmark (71.4), and that the two models place first and second on Hume AI's Overall Quality Index. It also reports top positions on the Voice Arena leaderboard for several major languages, and improvements in long-form content and two-speaker control over Gemini 3.1 Flash TTS. These are Google's figures from its own announcement.

Where you can use it

Developers can try the models in Google AI Studio. Google names Agora, LiveKit, Pipecat and Vercel as third-party integrations, and lists Figma, HeyGen, Linguana, Wondercraft, 99.co, Ollang, Spoken and Transforms.AI among the companies building with the models.

We couldn't confirm pricing from the announcement, so check the Gemini API documentation before budgeting for volume.

Safety and regional limits

All audio the models produce carries an imperceptible SynthID watermark so it can be detected as AI-generated, and Google says it uses C2PA credentials for provenance. Voice replication requires a recorded verbal consent.

Voice replication in AI Studio is not available in Illinois, Texas, the EEA, the UK, Switzerland or India.

Why it matters

Speech is turning into a normal building block for SaaS products: onboarding walkthroughs, support agents, product videos and in-app narration. A cheaper Flash-Lite tier suits high-volume features, while the fuller Flash model targets branded or character voices. Teams that ship audio to users in the regions above should plan for the replication limits and for consent handling, and anyone already using voice tools built on other engines now has a new option to test on cost and quality.