Gemini 3.8 Live Explained: What "Extended Thinking" Means When You Talk to Gemini, and Which Google AI Plan Gets It
What's in this guide
Google released two new voice models on September 15, 2026: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. If you have ever tapped the little Live button in the Gemini app, asked a question, and then sat through a long pause while it went off to think, this launch is aimed squarely at that pause.
The announcement is written for developers building call center bots, so it never quite gets around to the questions a normal person has. Which of the two models do I actually get? Do I need to pay? What does "extended thinking" mean when the thing is supposed to be talking to me in real time? So here is the plain English version, checked against Google's launch post, its developer documentation and its Google AI plans page on September 16.
The short answer
Gemini 3.8 Live Extended Thinking is the smarter of the two models, and it is the one that now powers Gemini Live, the voice mode inside the Gemini app. Google says it is rolling out to everyone, including people on the free plan, and I did not need to change a setting to get it.
The big change is that Gemini can now think and talk at the same time. When you ask something hard, it says a quick "let me check that," keeps the conversation going, does the reasoning in the background, and comes back with the answer. It can also run tasks (look things up, call other tools) while you carry on talking. In practice that means far fewer dead silences.
The plainer Gemini 3.8 Live model is the one behind Search Live, the voice mode inside Google Search. That is also free.
Where your subscription matters is Google Workspace. Talking to your inbox (Gmail Live) or your notes (Keep Live) needs at least Google AI Plus. Talking a document into existence (Docs Live) needs Google AI Pro or Ultra. Nothing about this launch requires the $99.99 or $199.99 Ultra tiers.
What Gemini 3.8 Live actually is (two models, not one)
The naming is confusing, so let me untangle it.
Gemini is Google's family of AI models. The regular text models, the ones that answer when you type, are things like Gemini 3.1 Pro. "Live" is Google's word for a different kind of model entirely: one that listens to audio and speaks audio directly, instead of transcribing your words to text, thinking in text, and reading the answer back out. Google's documentation calls this "audio-to-audio," and it is why Gemini Live can handle interruptions, tone and 97 languages mid-sentence without the robotic hand-off you hear in older voice assistants.
On September 15 Google shipped two of these:
- Gemini 3.8 Live is the workhorse. Google describes it as "built for scale and cost efficiency." It talks fluidly, understands what your camera sees in near real time, and can run tools in the background. Think of it as the fast, cheap one, tuned for products that millions of people will use at once.
- Gemini 3.8 Live Extended Thinking is the heavyweight. Same voice abilities, but with a much bigger reasoning engine bolted on behind the conversation. Google says it topped the Artificial Analysis Speech to Speech Quality Index with a score of 82.6, and it leads on benchmarks that measure whether a voice agent actually finishes multi-step tasks, not just whether it sounds nice.
Both models accept text, images, audio and video, and both can speak back. The odd part, and the good part for you, is which product got which model. The cheaper model went to Search Live. The expensive one went to Gemini Live in the app, and to the Workspace voice features, for everyone.
Why "thinking while talking" is the real change
Every "reasoning" or "thinking" model until now has had the same problem in voice: thinking takes time, and silence on a voice call feels broken after about two seconds. So voice modes either skipped the deep thinking and gave you shallow answers, or they made you wait.
Google's fix is what its documentation calls asynchronous reasoning. Here is what actually happens when you ask Gemini Live something hard now:
- It acknowledges you immediately, with a short verbal cue like "Let me check that" or "Give me a second."
- It starts reasoning in the background. If the task needs a tool, say a web search, a calendar lookup or a booking, it fires that off in the background too.
- It keeps talking. It can narrate its progress ("I found three options, comparing prices now") or answer a side question you throw in.
- When the background work is done, it folds the result into the conversation.
Developers can set how hard the model thinks (low, medium or high) for their own apps. Inside the Gemini app you do not get that dial. Google picks it for you.
One more detail worth knowing: the Extended Thinking model has "proactive audio" permanently switched on. That is Google's term for the model deciding on its own when to speak up, for example when it has finished a background task, rather than waiting for you to ask "are you done yet?"
If you read my Claude Fable 5.1 explainer or the GPT-6 Astra one, you know the text models have spent all year learning to think before they answer. This is that idea arriving in voice, minus the waiting.
Where each model shows up
This is the part the launch post buries. Here is where you will actually meet each model as a regular user.
| Product | Which model | Who gets it |
|---|---|---|
| Gemini Live (the voice button in the Gemini app) | 3.8 Live Extended Thinking | Everyone, including free |
| Search Live (voice mode in Google Search) | 3.8 Live | Everyone, including free |
| Gmail Live (talk to your inbox) | 3.8 Live Extended Thinking | Google AI Plus, Pro, Ultra |
| Keep Live (talk your notes into shape) | 3.8 Live Extended Thinking | Google AI Plus, Pro, Ultra |
| Docs Live (talk a document into a first draft) | 3.8 Live Extended Thinking | Google AI Pro, Ultra |
| Gemini API and AI Studio (for developers) | Both | Anyone with an API key |
Gmail Live, Docs Live and Keep Live are not brand new. Google launched them on September 3 for subscribers, running on an earlier Live model. What changed on September 15 is the brain behind them. Google Workspace business accounts are still listed as "coming soon" for the Extended Thinking upgrade, so if your employer pays for your Google account, you may be waiting a little longer.
Which Google AI plan gets what
Google's consumer subscriptions are called Google AI Plus, Google AI Pro and Google AI Ultra. Prices below are the US monthly prices on Google's plans page as of September 16, 2026. Plus and Pro double as Google One storage plans (400GB and 5TB), so if you pay Google for storage you may already have one without realising it.
| Plan | Price | Gemini Live with Extended Thinking | Search Live | Gmail Live and Keep Live | Docs Live |
|---|---|---|---|---|---|
| Free | $0 | Yes | Yes | No | No |
| Google AI Plus | $4.99 a month | Yes, higher limits | Yes | Yes | No |
| Google AI Pro | $19.99 a month | Yes, higher limits | Yes | Yes | Yes |
| Google AI Ultra | From $99.99 a month | Yes, highest limits | Yes | Yes | Yes |
A few things the table cannot show:
- Limits, not features, are what you pay for in the app. Free users get the same Extended Thinking model in Gemini Live. Paid plans get more usage before you hit a cap. Google does not publish the exact numbers, only that Plus is "2x," Pro is "4x," and Ultra goes up to "20x" a non-subscriber's access.
- The Docs Live gap is the only real paywall in this launch. If talking a document into a first draft is the thing you want, Plus will not do it. You need Pro at $19.99.
- Ultra buys you nothing extra here. The $99.99 and $199.99 Ultra tiers add Deep Think, Project Genie and much higher limits, but every voice feature in this post is available on Pro. Do not upgrade to Ultra for voice.
What I found on day one
I spent most of the evening of September 15 talking to my phone, which my family found hilarious. A few notes.
The pause really is mostly gone. I asked Gemini Live to plan a three day trip with a set budget, a dietary restriction and a specific arrival time, the kind of question the old version would have gone quiet on for eight or nine seconds. Instead it said "okay, let me work through that," asked me a clarifying question about hotel versus apartment while it was clearly still working, and then read out an itinerary about ten seconds in. The silence never went above a couple of seconds.
The camera trick is better than I expected. I pointed the phone at the back of my router and asked why my smart plug kept dropping off the network. It read the model number off the label, walked me through the 2.4GHz versus 5GHz band problem while I was still holding the phone up, and told me which setting to look for. That is the "visual grounding" Google talks about, and it is the first time it felt useful rather than a party trick.
Interrupting it works cleanly. I cut it off mid-sentence four or five times and it never lost the thread.
It is not perfect. Twice it announced "I've got that" and then read me something that did not quite match what I had asked, as if the background task had answered a slightly different question. And in a noisy kitchen it once responded to my partner instead of me. Those feel like early days problems rather than fundamental ones.
Docs Live, which I have on my Pro plan, is the sleeper hit. I rambled at it for four minutes about an article idea and it produced a structured outline with headings I would actually use. It pulled a date out of an email in my Gmail when I asked it to, after asking permission first. It is the closest thing yet to dictating to a human assistant who is also taking notes.
Should you care? A two minute decision
You do not need to buy anything to find out. Here is the quick version.
If you already use Gemini Live occasionally: open the app and try a hard question tonight. You already have the new model. The difference is obvious inside a minute.
If you have never used voice AI because it felt slow and dumb: this is the version to try. The thinking-while-talking change addresses the exact complaint that put most people off.
If you want to talk to your Gmail or your notes: Google AI Plus at $4.99 is the cheapest way in, and it comes with 400GB of storage, so if you are already paying Google for storage you may effectively have it.
If you want to dictate documents into shape: that is Docs Live, and it needs Google AI Pro at $19.99. Whether that is worth it depends on how much you already live in Google Docs. If you do, it probably is.
If you are on ChatGPT or Claude and wondering whether to switch: not for this alone. Voice is now Gemini's clearest advantage over the other two, but the reasons to pick an assistant are still the ones I laid out in my Claude vs ChatGPT vs Gemini comparison. Where your files live matters more than any single model.
The verdict
Gemini 3.8 Live Extended Thinking is the first voice AI I would recommend to someone who dislikes voice AI. The story is not the model. It is that Google removed the awkward silence that made talking to an AI feel like talking to a very smart person on a very bad phone line, and gave the fix to everyone for free.
For most people the plan question answers itself: stay on the free tier, try Gemini Live properly this week, and only consider Plus or Pro if you find yourself wanting to talk to your inbox or your documents. If you already pay for Pro because of the storage, you have the full set. Go and use it.
I will update this post when Google publishes exact usage limits or brings the Workspace features to business accounts.
Comments
Post a Comment