Gemini 3.8 Flash TTS vs Flash-Lite TTS: What Google Announced

Google’s Gemini 3.8 Flash TTS and Flash-Lite TTS generate spoken audio from text, but serve different production needs. Announced on September 23, 2026, Flash TTS emphasizes custom characters and creative direction; Flash-Lite TTS is designed for cost-efficient, high-volume work such as dubbing and voice agents. 427

Which model fits which job?

Gemini 3.8 Flash TTS is the creative option. Google says it can design a voice from a natural-language description of its role, accent and vocal characteristics, then follow performance cues for scripted scenes. That makes it the better fit when a distinctive voice and precise delivery matter most. 430

Gemini 3.8 Flash-Lite TTS is the scale-oriented option. Google positions it for high-volume dubbing, audio content and expressive voice agents, while retaining control over tone and pacing. The distinction is intended use—not that Flash-Lite is limited to flat, undirected speech. 2730

Voices, languages and performance controls

Google announced more than 2,000 ready-to-use voices, alongside the ability to create custom voices with prompts. It also described directing back-and-forth dialogue and specifying how individual lines should sound, including cues for tone, accents and pacing. 42829

Google describes voice creation across more than 100 languages and dialects. Third-party model listings give the more specific figures of 130 languages for Flash TTS and 101 for Flash-Lite TTS; those counts should be read as reported coverage rather than a guarantee that every voice or feature works in every language. 271718

Voice replication is separate from designing a new character voice: it uses a sample of an existing speaker. Reporting on the launch describes a roughly 30-second reference sample and a consent check before replication. 2223

Safeguards and benchmark claims

Google says voice replication includes consent verification. Launch reporting also describes SynthID watermarking on generated audio and C2PA credentials for replicated voices. These measures are safeguards, not a reason to assume every possible misuse has been eliminated. 2334

In results reported at launch, Flash TTS scored 71.4 and ranked first on Hume AI’s Voice Design Benchmark; another report lists 60.8 on an accent-modeling measure. Google also said Flash TTS and Flash-Lite TTS took the top two places on Hume AI’s Overall Quality Index. These are reported benchmark results, not an independent assessment of how either model will perform in a particular application. 1921

Where are the models available?

Google said both models began rolling out in Google AI Studio and the Gemini API on September 23. Its API release notes list the TTS models and a Voices endpoint as generally available. For consumer products, Google specified Flash TTS in Gemini Notebook and Flash-Lite TTS in Google Vids; it described access through Gemini Enterprise as coming soon. Product rollout should not be taken to mean every voice feature was available in every product at launch. 2932

Early integration reporting named platforms including Agora, LiveKit, Pipecat and Vercel, and companies exploring dubbing, localization or voice-agent uses, including Figma, HeyGen and Wondercraft. These examples indicate developer interest, not uniform availability across all integrations. 37