Google made two voice models for the Live API generally available. One is for fast conversation without pauses, the other reasons in the background while the conversation goes on.
- Gemini 3.8 Live is the default choice for most low-latency voice agents, with asynchronous function calling by default.
- Gemini 3.8 Live Extended Thinking reasons in the background during the live conversation.
- There is no price in the note, and the 97 languages come from Vercel's description.
Anyone who has talked to a voice assistant about something complicated knows the pause. You ask, and silence falls in which you do not know whether it heard you, whether it is thinking, or whether it simply hung. It is awkward.
Google's new models are an attempt to remove exactly that silence.
The important part is asynchronous function calling. In plain words: the assistant kicks off a task, checking an order or searching a calendar, and keeps talking to you while it waits for the answer. With synchronous calling, the conversation stands still until the tool returns something.
The second model goes further and thinks in parallel with speech. Vercel describes it like this: it can acknowledge that it heard the request and narrate how far it has got without interrupting the conversation. A good waiter does the same when an order runs late, and nobody thinks of him as slow.
The price of the new models is not in the note, and the 97 languages are Vercel's words, not Google's. Whether Bulgarian is among them and how it sounds will be found out by ear, not from a table.