Details
- Google DeepMind announces Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking as its best conversational AI models to date.
- The new Live models can talk, think, and execute tasks in the background, maintaining an uninterrupted conversational flow for users.
- Both variants offer upgraded reasoning capabilities designed to handle complex questions and multi-step tasks more reliably.
- The models support near real-time visual understanding, letting them ground conversations in live visual inputs such as camera feeds or shared screens.
- Gemini 3.8 Live automatically detects 97 languages, reducing setup friction for global users and enabling multilingual voice and chat experiences.
- Background tool calling allows the AI to trigger APIs, code execution, or other tools while continuing the dialogue, avoiding pauses or context switches.
- Google DeepMind highlights 3.8 Live Extended Thinking as suitable for demanding workflows, such as acting as a programming tutor for step-by-step guidance.
- The release positions the Live models as part of the broader Gemini 3.8 family, complementing earlier launches focused on coding, agents, and cybersecurity.
- Early benchmarks and product materials indicate Extended Thinking is optimized for deeper reasoning and agentic task completion compared with previous live dialogue models.
- Rollout spans developer and enterprise channels via Gemini APIs and Google AI tools, signaling a push toward more capable, always-on voice and multimodal agents.
Impact
By combining real-time voice, visual context, and background tool execution, Gemini 3.8 Live and Extended Thinking move Google’s stack closer to fully agentic AI experiences that can both converse and act. The models appear aimed at narrowing the gap with rival voice and assistant ecosystems from OpenAI and Anthropic, while giving enterprises a more scalable way to deploy multilingual, task-completing AI agents across customer support, developer tooling, and productivity workflows.
