Details
- Google for Developers announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, new audio-first models for natural, back-and-forth conversation.
- Gemini 3.8 Live targets scalable, cost-efficient voice agents that go beyond simple chat to understand the world, speak fluidly, reason through workflows, and execute actions.
- The model is designed for near real-time interactions, supporting asynchronous tool calling so background tasks can run while the conversation continues.
- Gemini 3.8 Live Extended Thinking focuses on complex, multi-step workflows, enabling deeper reasoning for requests that need longer, more deliberate processing.
- Extended Thinking lets agents reason and speak simultaneously, narrating their logic with natural verbal cues instead of using silent loading states.
- Both models are positioned to streamline how developers build intelligent voice applications, especially agents that must coordinate tools and workflows during live dialogue.
- The models are available now through the Gemini API via Google AI Studio, with availability coming soon to the Gemini Enterprise Agent Platform, expanding reach to enterprise use cases.
- Google highlights these releases as part of its broader push toward agentic, voice-first interfaces where AI can both converse and autonomously handle tasks in the background.
Impact
By pairing low-latency conversation with asynchronous tool use and a higher-reasoning variant, Google strengthens its position in voice-first agent platforms and narrows the gap with rivals like OpenAI and Anthropic that are also investing in real-time multimodal assistants. The Extended Thinking model, which keeps talking while it reasons, could accelerate adoption of AI voice agents in customer support, productivity and workflow automation by reducing friction from pauses and opaque processing, while the enterprise rollout signals Google’s intent to make these capabilities part of broader agent platforms rather than isolated APIs.
