Google Unleashes Gemini 3.8 Live: Bots That Think While Talking
Silicon Valley just upgraded voice assistants to do the unthinkable: hold a continuous speech conversation, think out loud, and query tools in the background without awkward silence.
New neural weights for Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking operate in native speech-to-speech mode, directly digesting video feeds and static images across more than 97 languages without relying on clunky middleman transcription pipelines.
The system leverages asynchronous tool calling, allowing the bot to fetch external API data behind the scenes while maintaining uninterrupted voice banter with callers who might never notice the background computational gymnastics.
Under the hood of the Live Extended Thinking tier, parallel reasoning enables the model to process complex multi-step problems while concurrently narrating its intermediate steps, a capability that netted it 82.6 points on the Speech-to-Speech Quality Index by Artificial Analysis.
Deployment began immediately across the Gemini API and Google AI Studio, alongside initial phases of feature integration into Google Search and Google Workspace.
Access economics run at $0.005 per minute for incoming audio streams and $0.018 per minute for synthesized voice output, scaling to $3 and $12 per million tokens on standard audio consumption tiers.
Awkward algorithmic lag once served as a comforting reminder of artificial limitations, but voice models capable of thinking aloud while running backend workflows quietly shift the dynamic toward conversational agents that simply leave humans with no room to interrupt.
Source: Google Blog
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.