Gemini 3.1 Flash Live: Real-Time Voice AI with 90.8% Benchmark Score

Google released Gemini 3.1 Flash Live on March 26, 2026 — an audio-to-audio model specifically designed for real-time voice applications. Unlike text-to-speech pipelines, Flash Live processes audio natively, enabling natural interruptions and low-latency responses.

Benchmark performance: 90.8% on ComplexFuncBench Audio — the highest score recorded for a voice AI model on multi-step function calling tasks. The model can handle complex voiced instructions like "Check my calendar, book the first free slot next week, and send a confirmation SMS" in a single conversational turn.

Use cases: Customer service phone bots, voice-driven CRM updates, real-time meeting assistance, and IVR replacements.

For businesses running phone-based customer interactions, Flash Live offers a significant step up from previous voice AI solutions with more natural conversation flow and reliable function execution.