Gemini 3.1 Flash Live: Building More Natural and Reliable Audio AI

Google has introduced Gemini 3.1 Flash Live, which takes real-time conversation capabilities to the next level. This model delivers the speed and natural rhythm needed for next-generation voice-centric AI, providing an intuitive experience for developers, enterprises, and everyday users alike.

Gemini 3.1 Flash Live is available through a variety of channels as follows.

• Developers: Available in preview via the Gemini Live API in Google AI Studio
• Enterprises: Available for use in Gemini Enterprise for Customer Experience
• Everyday users: Available via Search Live and Gemini Live

In terms of performance, the model has also shown notable results. It scored 90.8% on the ComplexFuncBench Audio benchmark, demonstrating complex multi-step function-calling capability, and achieved 36.1% in 'Thinking' mode on Scale AI's Audio MultiChallenge, showing strong reasoning ability even amid interruptions or hesitations that occur during real conversations.

The model's understanding of tone has also improved, allowing it to better recognize acoustic nuances such as pitch and pace. This enables it to dynamically adjust its responses in line with a user's frustration or confused expressions, resulting in even more natural conversations.