What's this about?
Google announced fast Live and deeper Extended Thinking versions for continuous voice and vision interaction. They can switch among 97 languages and call tools in the background without ending the conversation.
What exactly is a real-time voice agent?
Unlike dictation followed by a text chatbot, it listens and responds while the conversation continues, and may use a calendar, search or other tool mid-turn. That makes latency, interruptions and tool correctness matter together.
What's the community saying?
HN 322 points, 198 comments; independent Artificial Analysis ranked Extended Thinking first on its speech-to-speech agent index (82.6), but its human voice-preference score trailed the standard Live model.
- More natural multilingual conversation and asynchronous actions are promising.
- Rollout is uneven; some users could not access the announced model yet.
- Benchmarks and curated demos may not predict everyday voice quality or agent reliability.