Google is rolling out Gemini 3.8 Live voice models with real-time dialogue, visual grounding and parallel reasoning across apps and APIs.