Google launches Gemini 3.8 Live, voice models that think and talk at the same time

Google has released two new voice models for talking to Gemini. Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, announced on September 15, 2026, are designed to make spoken conversations with AI feel less like taking turns and more like working with someone: they keep talking while tasks run in the background, reason while they speak and switch languages mid-sentence. Google calls them its most advanced live dialogue models yet.

Two models, two jobs

• Gemini 3.8 Live is built for scale and cost efficiency. It combines conversational intelligence with fluid dialogue and visual grounding, and it powers Search Live.
• Gemini 3.8 Live Extended Thinking is built for complex tasks that need multi-step reasoning. It’s the model rolling out to Gemini Live in the Gemini app and to voice features in Gmail, Docs and Keep.

What’s new in the conversation

• Tasks in the background. Both models can call tools and APIs while the conversation continues, so Gemini can acknowledge a request and keep chatting while the work finishes.
• Thinking while talking. Extended Thinking reasons and speaks at the same time. It uses early cues such as “Let me check that…” and narrates its progress on multi-step tasks instead of going silent.
• 97 languages. Gemini 3.8 Live detects and switches between 97 supported languages in the middle of a conversation.
• Camera awareness. It processes visual input in near real time, so it can answer questions about what your camera shows.

In Google’s demos, the models guide a new employee through onboarding using what the camera sees, coordinate several bookings in one conversation and turn rough sketches and spoken feedback into working code.

Where you can use it

Gmail Live lets you search your inbox by talking, Docs Live drafts and edits documents from your voice, and Keep Live creates notes.

How it performs, according to Google

Google reports that Gemini 3.8 Live Extended Thinking takes first place overall on Artificial Analysis’ Speech to Speech Quality Index, with a score of 82.6. It scores 68.6% on the τ-Voice agentic benchmark, 35.1% on Sierra’s τ-Voice-banking test and 97.7% on Big Bench Audio. The standard Gemini 3.8 Live model placed second in the Speech Agent Arena, a ranking based on user preference. These are Google’s figures; independent comparisons with rival voice assistants will take longer.

For developers

Both models are available through the Gemini Live API. Voice platforms including Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel and Vision Agents handle the real-time audio streaming on top of it, and Google names Salesforce, Genspark and Lumeris among its early partners. Google hasn’t published new pricing for the models in its announcement.

Watermarked audio

All audio generated by Google’s AI products is watermarked with SynthID, an imperceptible marker built into the sound itself to help keep AI-generated audio detectable.

A busy week for voice assistants

The launch came a day after Apple released Siri AI with iOS 27, and it follows Gemini 3.8 Flash, which Google calls its most intelligent workhorse model, released on September 2. On Android, Gemini is also replacing Google Assistant on eligible phones. For how Gemini compares with OpenAI’s assistant, including voice, read Gemini vs ChatGPT, and see our Gemini profile for plans and models. OpenAI’s rival voice models are covered in our guide to ChatGPT Voice, and Gmail Live is one of the tools in our roundup of the best AI email assistants. For Gemini in the home, see Google Home Premium explained.