Somewhere between "ask" and "answer," most AI still makes you wait. Google just closed that gap, and rolled it straight into Gmail and Docs while nobody was watching.

Here's what that actually means for how you work.

It's not a demo. It's already sitting inside Gmail, Docs, and Keep.

There's a specific kind of silence you get used to with voice assistants. You ask something with a bit of nuance to it, and it goes quiet for a second while it figures out what to say. Most of us have learned to just wait it out.

Google's latest release is aimed squarely at that pause. It just put out two new voice models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking.

The pitch isn't "smarter answers." It's an AI that keeps talking while it's still working things out, and it's already showing up in apps a lot of us use every day without thinking twice about it.

Here's the thing Google got right this time: it didn't try to cram everything into one model.

Think of Gemini 3.8 Live as the quick, chatty one. You ask something simple, it answers fast, and it can even glance at what you're showing it and switch languages mid-sentence if you do.

Gemini 3.8 Live Extended Thinking is the one you'd call in for the harder stuff. The kind of question that actually needs a bit of thought before answering. So Google routes the easy asks to the fast model, and saves this one for when things get complicated.

Most voice AI works like this: you ask, it goes silent, then it answers.

Extended Thinking skips that gap.

It talks while it thinks, using little verbal placeholders like "let me check that" to bridge the silence. Then it narrates each step out loud as it finishes.

It can also run background tasks and API calls mid-conversation. So it can acknowledge what you asked for out loud, while the actual work is still happening behind the scenes.

This isn't tucked away in a developer console somewhere. It's showing up in tools a lot of us already use for daily writing and editing:

If you're on AI Pro or Ultra, you get Extended Thinking across all three. Everyone else gets the standard Live model through Search Live.

Google's own launch post cites some strong scores for Extended Thinking, including the top spot on Artificial Analysis' Speech to Speech Quality Index and a 97.7% on Big Bench Audio. Worth flagging: these numbers come from Google, not from an outside audit.

Good signal, not gospel.

A couple of other things to know before you build a habit around this. There's no open-weights version, so it only runs inside Google's own ecosystem.

Pricing is metered per minute of audio, with extra charges for reasoning tokens on Extended Thinking. A heavy daily habit could get pricier than it first looks.

If you've ever dictated a draft on a walk, or talked through an email reply instead of typing it out, this closes a gap that's been there for a while. Most voice tools either transcribe what you say or answer a question.

This one can hold an actual conversation and hand you a finished draft, without breaking stride to "think."

That's the real story here, not the benchmarks. Google is treating voice as something you produce work through, not just something you ask questions with.

If that sticks, talking your way through a first draft might stop feeling like a gimmick.