Gemini 3.8 Live splits fast dialogue from extended reasoning

Google announced Gemini 3.8 Live and Live Extended Thinking on 15 September, with the post updated on 17 September. The standard model targets efficient dialogue and visual grounding, while Extended Thinking is positioned for harder tasks requiring multiple reasoning steps. Access spans developer, enterprise and consumer experiences.

The design separates conversational continuity from background work. An agent can keep responding while tools or more involved reasoning proceed, rather than treating every request as a single spoken answer. Google reports stronger voice and task-completion evaluations, but the higher-effort model still needs an application to define what it may do. A voice interface can make a workflow feel continuous even when its underlying actions retain separate latency, permissions and verification requirements.