Google DeepMind Introduces Gemini 3.8 Live Models for Real-Time Voice Agents

Google DeepMind has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, positioning the pair as its strongest conversational AI models for voice-first assistants that can keep a conversation moving while reasoning and handling work in the background. The release expands the Gemini Audio lineup from low-latency dialogue toward agentic, multimodal interaction: Gemini 3.8 Live is described as best for near real-time voice interfaces, while Gemini 3.8 Live Extended Thinking is aimed at complex reasoning and high-complexity tasks that can think and narrate progress in near real time.