Google unveils new AI model for enhanced voice interactions
What's the story
Google has unveiled its latest AI model, the Gemini 3.1 Flash Live, to improve voice-based interactions by reducing latency and improving accuracy. The new system can handle real-time dialogues more effectively, providing smoother back-and-forth conversations across various platforms including Search, Gemini Live, and developer platforms. The model's enhanced tonal understanding allows it to detect variations in pitch, pace, and user intent for more context-aware responses.
Developer benefits
Improved task execution and reasoning
The Gemini 3.1 Flash Live model also offers improved task execution and reasoning in voice-based applications for developers. It has shown higher performance on benchmarks that measure multi-step instruction handling and complex function execution in audio environments. The model is available across multiple Google platforms, including through the Gemini Live API in Google AI Studio and via Gemini Enterprise for customer experience use cases.
Global accessibility
Model integrated into Gemini Live and Search Live
The Gemini 3.1 Flash Live model is integrated into Gemini Live and Search Live for end users. Google has expanded Search Live to over 200 countries and territories, enabling real-time interaction with search using voice and multimodal inputs. The model supports voice-first applications such as customer service systems, virtual assistants, and real-time troubleshooting tools in noisy environments.
Responsible AI
Google embeds SynthID watermarking in audio generated using Gemini
Google has also embedded SynthID watermarking in all audio generated using the Gemini 3.1 Flash Live model. This imperceptible signal helps identify AI-generated content and reduce misuse. With this update, Google is taking a major leap toward real-time, conversational AI across its products and services while ensuring responsible use of technology.