Google's new voice model switches language mid-conversation
On 15 September Google introduced two voice models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The first is built for continuous conversation at a low price. The second is built for tasks that need several steps of reasoning. Both hold a real-time voice dialogue, run actions in the background without stopping the conversation, and process visual input in near real time, so you can point a camera and talk about what it sees while it sees it.
The number that reaches a small Israeli business is in the announcement itself. The model detects 97 supported languages on its own and moves between them in the middle of a conversation, without the user choosing a language first.
Who can use it today
The split between channels decides who can touch this now. Developers get both models in the Gemini API and in Google AI Studio. Organisations get them in Gemini Enterprise as a closed preview, and the rollout to Gemini Enterprise for Customer Experience and to businesses on Google Workspace was named as the next step, with no date.
A regular user gets less, though not nothing. Gemini 3.8 Live is open to everyone in Search Live. Extended Thinking runs in the Gemini app's live conversation, in Docs for Google AI Pro and Ultra subscribers through Workspace, and in Gmail and Keep for anyone subscribed to Google AI.
A business owner who wants to try it today does it from the phone, in the Gemini app's live conversation. The app supports Hebrew and Israel is on the list of supported countries. Full Hebrew support in live conversation arrived in January 2025, free users included.
97 languages and switching mid-sentence
Detecting the language while someone speaks sounds like a technical footnote, and it is the detail that decides what the tool is worth to a business that takes enquiries in several languages. A customer who opens in Hebrew, drops in an English term and moves to Russian does not break the conversation. The same holds in the other direction, when the owner is drafting a reply to an English-speaking customer and wants to switch language without starting over.
Anyone who invoices clients abroad already knows the document side of this, which is a separate matter from the conversation. What has to appear on a document going to a foreign client is set out in the guide to invoicing in English.
Uses that fit a one-person business
Visual input in real time opens uses that were awkward in a text chat. You can point a camera at an item in storage and ask about it as you go, or run through a form and ask what a particular clause says before filling it in. Voice also suits drafting a reply to a customer while driving, and rehearsing a sales call before the meeting.
The thing to avoid is reading out identifying customer details for no reason. A voice conversation feels more private than a screen. Everything said goes to an outside provider's server, and customer details that land there do not come back.
Where it stops
The model talks, sees and runs actions in the background. It does not issue an invoice, request an allocation number or file VAT. Everything that touches reporting stays in the system that keeps the books, and the voice conversation is a layer above it.
The second gap is between the announcement and the availability. The part meant for businesses on Workspace has not opened yet, and Gemini Enterprise is still a closed preview. A business that wants to build automated customer service on this waits for the next stage, or works against the API and pays per use. Anyone who just wants a voice assistant that understands Hebrew can open the app today.