Gemini 3.1 Flash TTS Preview

Overview. Gemini 3.1 Flash TTS Preview is Google's dedicated text-to-speech model designed to convert text prompts directly into synthetic speech. Built on the Flash architecture for responsive processing, it supports audio generation across more than 70 languages with steerable styling and expressive inline tags. It operates as a preview endpoint scheduled for deprecation in November 2026 in favor of Google's Gemini 3.8 speech synthesis series.

Key strengths. The model generates 24 kHz 16-bit mono audio, outputting standard WAV files for unary calls and chunked PCM streams for real-time playback. It handles up to 8,192 text input tokens and can synthesize long audio sequences of up to 16,384 output tokens. Expressive inline tags allow developers to direct tone, pace, and vocal inflection without separate prosody models.

When to use it. Use this model for legacy text-to-speech integrations that require low-latency streaming audio from raw text input. Because Google has deprecated this preview model ahead of a November 17, 2026 retirement date, teams maintaining active implementations should transition to Gemini 3.8 Flash TTS or Gemini 3.8 Flash-Lite TTS. For greenfield audio workflows on the Platform, select the newer 3.8 series models directly.

Limitations. The model is strictly restricted to text-to-speech synthesis and cannot output text, process image or audio inputs, call external tools, or support context caching. It lacks reasoning and code execution capabilities. The preview status also comes with an active deprecation schedule leading to complete service retirement in late 2026.

Sovereignty note. Inference runs on Google infrastructure in the United States, exposing data flows to the US CLOUD Act. Organizations requiring EU Sovereign operational isolation should evaluate Google cloud processing terms against their regulatory data residency requirements.