Google has unveiled two new real-time conversational models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The former is designed for cost-efficient large-scale deployment, while the latter focuses on complex multi-step reasoning tasks.
Gemini 3.8 Live Extended Thinking achieved an 82.6% score on the Artificial Analysis voice-to-voice quality index and a 68.6% completion rate in τ-Voice agent tasks. In the Big Bench Audio benchmark, it scored 97.7%.
Gemini 3.8 Live supports real-time visual input processing and automatic switching between 97 languages. It allows for background API calls without interrupting the flow of conversation.
Developers can access the new models via the Gemini API and Google AI Studio. Integration is supported on platforms such as Agora, Fishjam, LiveKit, Pipecat, Vercel, and Vision Agents.
Google has implemented SynthID invisible watermarking for all AI-generated audio to identify content and mitigate misinformation. Enterprise users can access the models through Gemini Enterprise and Google Workspace.
