Google has released Gemini 3.1 Flash Live, its latest real-time voice and audio model. Valeria Wu and Yifan Ding write in the Google Blog that the model offers faster responses and improved natural conversation compared to its predecessor.
The model is available in several Google products. Developers can access it via the Gemini Live API in Google AI Studio. Enterprises can use it through Gemini Enterprise for Customer Experience. General users encounter it in Gemini Live and Search Live.
Google highlights improved performance on several benchmarks. On ComplexFuncBench Audio, which tests multi-step tasks, the model scores 90.8 percent. On Scale AI’s Audio MultiChallenge, which tests reasoning in realistic audio conditions, it reaches 36.1 percent with its “thinking” mode enabled.
The model also better recognises acoustic cues such as pitch and speaking pace. It can adjust its tone when a user sounds frustrated or confused.
In Gemini Live, the model doubles the length of conversation it can follow, which helps during extended discussions. It also responds faster than the previous version.
Google is also expanding Search Live to more than 200 countries and territories this week, supported by the model’s multilingual capabilities.
All audio generated by Gemini 3.1 Flash Live includes an imperceptible watermark from Google’s SynthID system to identify AI-generated content.
