Google is giving Gemini Live and Google Search Live a serious audio upgrade with the rollout of Gemini 3.1 Flash Live, a new real-time voice model that aims to make talking to AI feel more like talking to a person.

At a high level, Gemini 3.1 Flash Live is an audio‑to‑audio model built for low‑latency, live conversations, not just dictation or simple voice commands. It’s designed to respond quickly, handle back‑and‑forth dialogue, and keep track of the context so you don’t have to repeat what you said a minute ago. Google says it’s now powering Gemini Live in the Gemini app and Search Live, so the same model is behind both your chatty AI assistant and voice‑driven search sessions.

In practical terms, you should notice that voice chats in Gemini Live feel smoother: fewer awkward pauses, faster responses, and better ability to follow long, rambling questions or brainstorming sessions. Under the hood, Google has focused on “tonal understanding,” meaning the model is better at picking up cues like pitch, pace, and even frustration in your voice, and then adjusting its tone and answer length to match the moment. It’s also better at separating your voice from background noise like traffic or TV, which matters if you’re talking to it on the go.

Search Live is getting a boost, too. Backed by Flash Live, it can now offer more natural, real‑time multimodal conversations — think asking a follow‑up question with your voice while pointing your camera at something, and having Search keep the thread going instead of treating each query as a fresh start. Google is using this model to take Search Live global, rolling it out to more than 200 countries and territories in the languages where AI features in Search are already available.

For developers, Gemini 3.1 Flash Live is available in preview via the Gemini Live API in Google AI Studio, which means you can start building your own real‑time voice agents on top of the same tech that powers Gemini Live. The model supports function calling, so it can listen to a user, reason through multi‑step tasks, and call tools or APIs as needed, instead of just answering questions. Early benchmarks show strong performance on complex audio tasks, including multi‑step function calling and long‑horizon reasoning in messy, real‑world audio with interruptions. Google is already pitching this to enterprises for customer experience use cases, where natural, low‑latency voice bots are becoming a big differentiator.

All of this fits into Google’s broader push to make Gemini feel like a single, always‑available AI that you can talk to anywhere: in the Gemini app, inside Search, and inside third‑party apps via APIs. Flash Live is essentially the audio engine behind that vision — moving beyond “press the mic, wait for text” toward continuous, real‑time conversations that blur the line between assistant, search, and agent.