Gemini 3.1 Flash Live: Google's latest AI audio model

Google introduced Gemini 3.1 Flash Live, its newest audio and voice model for real-time dialogue. The model is available in preview for developers through the Gemini Live API in Google AI Studio. Enterprises can access it in Gemini Enterprise for Customer Experience. It is also available to users through Search Live and Gemini Live. Google reported that the model achieved a score of 90.8 percent on ComplexFuncBench Audio, a benchmark for multi-step function calling. It scored 36.1 percent on Scale AI's Audio MultiChallenge with thinking enabled. The model shows improved tonal understanding and better recognition of acoustic nuances such as pitch and pace compared with 2.5 Flash Native Audio. It also adjusts responses more effectively to expressions of frustration or confusion. Verizon, LiveKit and The Home Depot provided positive feedback on its natural conversation capabilities in their workflows. In Gemini Live and Search Live the model produces faster responses and maintains conversation threads for twice as long as the prior version. The model is multilingual and supports the expansion of Search Live to more than 200 countries and territories. All audio output includes a SynthID watermark.