Gemini 3.8 Live Explained: Google's Two New Voice Models, and Which One You Get

Google released two new voice models on September 15, 2026: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. You will never pick either one from a menu. They are the engines behind the talk-out-loud buttons you already have, and the only thing you need from this page is which of your apps got which model, and whether any of it costs money. The short version: the better model went into the free Gemini app, and the cheaper one went into Google Search.

Key Takeaways

• Two models, launched together on September 15, 2026. Google describes plain 3.8 Live as "built for scale and cost efficiency" and 3.8 Live Extended Thinking as "built for high-complexity tasks, with increased intelligence and multi-step reasoning."
• Gemini Live gets the smarter one, for everyone. Google's rollout list puts 3.8 Live Extended Thinking in Gemini Live under the heading "For everyone." Gemini Live is on the free plan on Google's subscriptions page.
• Search Live gets the cheaper one, also for everyone. Google's help article for Search Live names no plan at all: it is available "in all regions and languages where AI Mode is available," in the Google app on Android and iOS.
• The Gmail, Docs and Keep versions stay paid. Extended Thinking reaches Docs for "Google AI Pro and Ultra subscribers" and Gmail and Keep for "all Google AI subscribers." Those plans start at $4.99 a month. See Gmail Live, Docs Live and Keep Live.
• The headline change is that it stops stalling. Google says the models execute tools and API calls "in the background while continuing the conversation," so the assistant can acknowledge a request and keep talking while the task finishes.
• It switches languages mid-sentence. Google says 3.8 Live "automatically detects and transitions between 97 supported languages mid-conversation." It has not published the 97.
• Nothing to install, no price change, no setting. There is no model picker, no release note, and no announcement that any plan or price moved.

Which Product Gets Which Model

Here is Google's own rollout list, reorganised around the question an ordinary person asks, which is "do I have this?" Everything in the consumer rows is Google's wording from the announcement.

The row worth staring at is the second one. Google put the model it calls cost-efficient into Search and the model it calls high-complexity into the free Gemini app. If you have both apps on your phone and you want the more capable voice assistant, open Gemini rather than the Google app.

What Actually Changes When You Talk To It

Three things, and all three are about the conversation not breaking.

It keeps talking while it works. This is the change you will notice first. Google says the models "execute tools and API calls in the background while continuing the conversation, so the model can acknowledge requests and keep chatting while tasks finish in the background." The old behaviour, where you ask for something and the voice goes silent until it is done, is what this replaces.

Extended Thinking narrates instead of pausing. Google says it "reasons and speaks simultaneously," using "early verbal cues like 'Let me check that...' to acknowledge prompts naturally, and live progress narration to walk users through multi-step background tasks as they progress." In plain terms: on a request that takes a while, it tells you what it is doing rather than leaving dead air.

It follows you between languages. Google says 3.8 Live "automatically detects and transitions between 97 supported languages mid-conversation." That is a genuine change for bilingual households, but treat the number carefully: Google has not published which 97, and every product this model sits inside applies its own language rules on top. Gmail Live and Docs Live are English-only whatever the model can do.

It also reads your camera faster. Google says 3.8 Live "processes visual inputs in near real-time," which is the part that matters if you use Gemini Live or Search Live by pointing your phone at something.

Search Live: The Gates Google Does Not Put In The Announcement

The announcement says "For everyone: In Search Live" and leaves it there. Google's Search Live help article adds the limits, and there are four.

Two of those are worth a sentence each. The region rule is not a short list: Google's AI Mode article enumerates more than 240 countries and territories and roughly 90 languages, so most readers are inside it. And Search Live keeps a record: Google says your conversations appear as transcripts in your AI Mode history, and you can delete them there, thread by thread.

One more practical note from the same article. Search Live carries on running when you leave the app or lock your screen, controlled from a notification, unless you turn "Continue in background" off. Your camera does not: Google says it switches off when you background the app or lock the phone and does not turn itself back on.

Is This Better Than What You Had?

On Google's own numbers, yes, though the numbers are aimed at developers rather than you. Google says 3.8 Live Extended Thinking took "the #1 overall spot on Artificial Analysis' Speech to Speech Quality Index (82.6)," leads "in agentic task completion with 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark," and scored "97.7% on Big Bench Audio." Plain 3.8 Live took second place in the Speech Agent Arena.

What none of that tells you is whether ordinary conversation feels different, and Google publishes no consumer-facing comparison. The claims that are checkable against your own use are the three in the section above: background tasks, spoken progress, and language switching.

Two things have not changed and are worth stating because Google did not. The features that were off in Gemini Live are still off: Google's Gemini Live overview still says Gems, Notebooks, Omni and Lyria do not work during a voice session, and a new model underneath does not change that. And Gemini Live is still phone and tablet only, per its help article. There is no desktop version of either Live feature.

Should You Do Anything

Do nothing if you already use Gemini Live or the Live button in the Google app. The upgrade is server-side and arrives on its own. There is no update to install and no switch to find.

Open the Gemini app instead of the Google app if you want the better of the two models. Google's rollout list is explicit: Extended Thinking went to Gemini Live, plain 3.8 Live went to Search Live, and both are free.

Do not start paying for this. Nothing in the announcement moves a price or adds a feature to a paid tier that was not already there. The Gmail, Docs and Keep voice features cost what they cost since September 3; the models behind them got better at no extra charge. If you are weighing a subscription for other reasons, our guide to which AI chatbot deserves your money covers the wider question, and the best free AI chatbot in 2026 covers the case for paying nothing.

Check your settings if you did not know Search Live kept transcripts. They live in AI Mode history in the Google app, and you can delete them individually or clear the lot.