Google DeepMind releases Gemini 3.8 Live and Live Extended Thinking

Google DeepMind announced Gemini 3.8 Live and 3.8 Live Extended Thinking, its new state-of-the-art live audio models. The models support 97 languages with seamless switching, near real-time visual understanding, and background tool calling. The Extended Thinking variant adds increased performance and precision for complex tasks. The models are available in Gemini Live and via the Gemini API.
Key insights
New flagship live audio models released
Google DeepMind officially released Gemini 3.8 Live and 3.8 Live Extended Thinking as its new state-of-the-art live audio models. The release was announced on September 15, 2026, with the models positioned as the company's best conversational AI. The launch includes availability in the Gemini app and through the Gemini API for developers.
Multilingual and multimodal capabilities highlighted
The new models feature support for 97 languages with automatic detection and seamless switching between them. They also include near real-time visual understanding, allowing the AI to process visual information during conversations. These features are designed to make the models more versatile for global users and complex interactions.
Extended Thinking variant targets complex tasks
The 3.8 Live Extended Thinking model adds increased performance and precision for the most difficult tasks. It narrates task progress to keep the conversation going while working through problems. A demonstration showed the model acting as a programming tutor, showcasing its ability to handle educational and technical scenarios.