Gemini 3.8 Live Is Here: Google’s AI Can Now Talk, See, and Think at the Same Time

If you asked Google Assistant a question last year, you know the drill: you talk, it thinks, then it answers, one turn at a time. Gemini 3.8 Live throws that pattern out. Released on September 15, 2026, it’s Google’s new voice AI that can listen, glance at what your camera sees, reason through a multi-step task, and keep chatting with you, all while it’s still quietly working in the background.

• Launch date: September 15, 2026, rolling out immediately
• Two models: Gemini 3.8 Live (fast and cheap) and Gemini 3.8 Live Extended Thinking (deeper reasoning)
• Languages: Switches fluently between 97 supported languages mid-conversation
• Benchmark score: 82.6 on Artificial Analysis’s Speech-to-Speech Quality Index, currently the top score recorded
• Where it lives: Search Live and Gemini Live today, plus Workspace, the Gemini API, and Google AI Studio

What Makes Gemini 3.8 Live Actually Different

Older voice assistants work like a walkie-talkie. You speak, there’s a pause, then you get an answer. Nothing happens in between. That gap is exactly what Google built Gemini 3.8 Live to close.

The model processes audio and video input in near real time, so it can react to what it sees through your phone’s camera while you’re still mid-sentence. Google says it now throws in natural verbal cues, things like “let me check that,” while it works. It’s a small touch, but it’s the difference between a tool that feels responsive and one that feels like it froze.

The bigger shift is that the model can run tools and hit APIs without stopping the conversation. Ask it to check flight prices, and it can keep talking about your trip while the search happens quietly underneath.

Extended Thinking Is the Version Actually Worth Watching

Gemini 3.8 Live is built for speed and low cost. Gemini 3.8 Live Extended Thinking is built for jobs that need a few steps to finish, and it’s the more interesting release of the two.

Extended Thinking only makes asynchronous tool calls, meaning it never has to stop and wait on an API response before it can keep talking. Picture asking it to book a dinner reservation: it can confirm times, cross-check your calendar, and text a friend, narrating its progress the whole way, instead of going silent for ten seconds and then dumping a wall of text on you.

That async-only design is a deliberate trade-off. It sacrifices some raw response speed for a conversation that never feels like it’s stalling out.

How the Two Models Compare on the Numbers

Google published a handful of benchmark results alongside the launch, and the gap between the two models is worth a closer look before you pick one for a project.

Neither model is universally “better.” The plain Live model is the one you’d pick for a customer support line handling thousands of quick calls, while Extended Thinking fits jobs that actually need to think between steps.

Where You Can Actually Use It Right Now

This isn’t a research demo sitting behind a waitlist. Google pushed it live across several products at once, though not every surface got the same model.

If you’ve already moved off Google Assistant and onto Gemini, you don’t need to do anything to try this. It’s just quietly running under the hood now.

What It Costs If You’re Building With It

For everyday users, this is free within whatever Gemini plan or app you’re already using. Developers pay per token, though, and the pricing tells you who this is really aimed at.

Audio input runs $3 per million tokens, or roughly half a cent per minute. Audio output costs $12 per million tokens, closer to two cents per minute. According to TechRepublic’s breakdown of the launch, both models can still hallucinate and occasionally time out, so Google isn’t pretending this is a finished product yet.

This release also lands alongside Gemini 3.8 Flash Cyber, Google’s security-focused model from the same generation, suggesting the 3.8 line is being split by use case rather than shipped as one do-everything model.

Frequently Asked Questions

What is Gemini 3.8 Live?

It’s Google’s real-time voice AI model, launched September 15, 2026, that can hear, see through your camera, and take actions in the background without breaking off the conversation.

How is Gemini 3.8 Live Extended Thinking different?

Extended Thinking is the slower, smarter sibling. It’s built for multi-step tasks like bookings or research, and it only uses background tool calls so it can keep talking while it works.

Is Gemini 3.8 Live free to use?

Yes, if you’re using it through Search, the Gemini app, or Workspace on an existing plan. Developers building their own apps pay per token through the Gemini API.

Which languages does it support?

Google says it automatically detects and switches between 97 supported languages mid-conversation, without you needing to change any settings.

Can I use it on Windows or inside Gmail?

Extended Thinking is rolling out inside Gmail and Keep for AI subscribers now. On desktop, it also runs through the Gemini app for Windows, which already supports Gemini Live conversations.

Our Take

The overlap-while-you-talk trick sounds like a small UX tweak until you actually use it, and then the old stop-and-wait assistants feel broken by comparison. Extended Thinking is the one to watch: it’s not the flashiest benchmark winner, but background tool use is what turns a voice assistant into something you’d actually trust to handle a booking. If you’re already inside the Gemini ecosystem, there’s nothing to install, and that alone makes this worth a few minutes of your time this week.