Google Gemini 3.1 Flash Live: Real-Time Multimodal Voice Model — 90+ Languages, 90.8% Function-Calling Accuracy

Google DeepMind·medium signal

Google released Gemini 3.1 Flash Live, their highest-quality audio model for real-time dialogue. Supports 90+ languages, filters background noise, follows conversation threads 2x longer than previous model, scores 90.8% on ComplexFuncBench Audio for multi-step function calling with constraints. All audio is watermarked to prevent misinformation. Available via Gemini Live API in Google AI Studio and Search Live in 200+ countries. The 90.8% function-calling accuracy means agents can reliably trigger tool use mid-conversation.