Google has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, new models designed for real-time voice conversations and more complex tasks.
The models aim to make conversations with AI more natural. Gemini can adapt when users interrupt it, supports 97 languages, and can analyze objects shown through the camera. For example, users can show a broken bicycle part through the camera, and Gemini can identify the problem and explain what to do next.
The Extended Thinking version is designed to handle complex, multi-step tasks while continuing the conversation. For instance, it can work on an event plan without interrupting the ongoing dialogue.
Gemini 3.8 Live Extended Thinking scored 82.6 points and took first place in Artificial Analysis’ Speech-to-Speech Quality Index, ahead of voice models from OpenAI and xAI. Gemini 3.8 Live is currently available to developers through the Gemini API and Google AI Studio.
