Google Gemini 3.8 Live

Google Gemini 3.8 Live is a native speech-to-speech model released September 15, 2026, designed for real-time voice and vision interactions over a stateful WebSocket connection. It accepts 16kHz PCM audio, images, text, and video as input and returns 24kHz PCM audio. The model covers 97 languages with consistent accent handling across all supported locales, the broadest language coverage among native speech-to-speech models available as of late 2026. Pricing is approximately $0.023 per minute combining audio input at $0.005/min and audio output at $0.018/min. The 131K token context window supports extended dialogue sessions without context truncation. Asynchronous function calling allows the model to execute API requests and tool calls in the background while streaming audio continues, reducing perceived response delays in tool-heavy agent flows. Both free and paid tiers are available. Gemini 3.8 Live is best positioned for teams building multilingual voice products where breadth of language support and low per-minute cost are primary requirements. Python and JavaScript SDKs plus integrations with LiveKit, Pipecat, Agora, and LangChain reduce integration friction. Session limits of 15 minutes for audio-only connections can be extended via session management techniques documented in the Google AI SDK.