This Gemini model talks while it thinks

I gave Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking the same prompt: draw me a pelican riding a bicycle up a mountain. Gemini 3.8 Live made the tool call fast. Extended Thinking took longer, but it kept talking the whole time, telling me what it was drawing before it sent the SVG. I think it drew the better picture too. (Waits play at 3x.) One thing to know if you build with it: on Extended Thinking, "turn complete" doesn't mean the interaction is done. More audio and the tool call can still arrive after it. The LiveKit Google plugin handles that for you, so your agent code stays the same for both models. Simple tool input where speed matters? Start with Gemini 3.8 Live. Something more complex? Test Extended Thinking and decide if the better output is worth the wait. Same agent, one string to change. #gemini #voiceai #livekit