Gemini 3.8 Flash TTS

Google's text-to-speech models of the 3.8 generation, released September 23, 2026 as two API models: Gemini 3.8 Flash TTS (gemini-3.8-flash-tts), built for voice design and directed performance, and Gemini 3.8 Flash-Lite TTS (gemini-3.8-flash-lite-tts), built for high-volume, low-latency speech and the recommended replacement for gemini-3.1-flash-tts-preview. The shared Gemini 3.8 Audio model card says both are based on Gemini 3 Pro, accept up to 8K tokens of text and return up to 64K audio tokens. Flash TTS designs new voices from natural-language prompts (role, accent, voice characteristics), offers a library of more than 2,000 voices, replicates a voice from a 30-second sample only after a verbal consent recording from the voice's owner, stages two-speaker scenes from one script, and follows inline vocal events such as <laugh> or <sigh>. The API documentation lists 130 languages for Flash TTS and 101 for Flash-Lite. Every generated clip carries a SynthID watermark, and voice replication adds C2PA credentials.

Google's evaluation sheet (September 2026, production checkpoints, single attempts) puts Flash TTS at 0.920 overall on Hume AI's text-to-speech quality benchmark (reliability times expressiveness) and Flash-Lite at 0.914, ahead of Cartesia Sonic 3.6 (0.840), Gemini 3.1 Flash TTS (0.783), ElevenLabs v3 conversational (0.769), OpenAI's gpt-4o-mini-tts (0.740) and ElevenLabs v3 (0.706). On Hume's voice-design leaderboard Flash TTS rates 71.4 overall in English, against 70.8 for ElevenLabs Voice Design v3 and 69.8 for Inworld, and 60.8 on accents against 45.4, while ElevenLabs leads on voice qualities (76.6 against 74.6). In Voice Arena's blind pairwise Elo ratings, Flash TTS leads the compared models in Japanese (1232), Arabic MSA (1204), Mexican Spanish (1152) and Hindi (1106), and Flash-Lite leads in Vietnamese (1156), Brazilian Portuguese (1134) and English (1087, against 1068 for Cartesia Sonic 3.6 and 1061 for Flash TTS). On Artificial Analysis's independent Text to Speech Arena on September 30, 2026, Flash TTS ranked third with an Elo of 1268, behind Eleven v4 (1316) and Sonic 3.6 (1275); Flash-Lite TTS ranked seventh (1240) and Gemini 3.1 Flash TTS twelfth (1204).

Gemini API pricing through December 31, 2026 is $0.50 per million text input tokens for both models and $9.00 (Flash) or $6.00 (Flash-Lite) per million audio output tokens, about $0.00225 and $0.0015 per 10 seconds of audio; all prices double on January 1, 2027. Flash TTS also runs in Gemini Notebook and Flash-Lite TTS in Google Vids, with Gemini Enterprise access announced as coming soon. The model card gives a January 2025 knowledge cutoff and, relying on Gemini 3.7 Flash's frontier-safety evaluations, judges the audio models unlikely to reach any tracked or critical capability level. Proprietary; parameter counts and architecture details beyond the Gemini 3 Pro base are undisclosed.