Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
deepmind ·
Google has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, marking a significant advancement in live dialogue models for voice interactions. These new capabilities aim to make conversations with AI more natural, fluid, and intelligent by handling complex reasoning and real-time visual context without interrupting the user. Gemini 3.8 Live offers fast, cost-efficient performance with support for 97 languages and background tool execution, while the Extended Thinking variant focuses on high-complexity tasks using multi-step reasoning. The models have achieved top rankings in benchmarks such as Artificial Analysis' Speech to Speech Quality Index and ServiceNow’s EVA-Bench. Users can access these features through the Gemini API, Google Workspace applications like Docs and Gmail, and Search Live. For developers, integration is supported via platforms like LangChain and Vercel to build production-ready voice agents. Additionally, all audio generated by these models includes SynthID watermarking to ensure transparency and help prevent misinformation.
deepmind