What is Gemini 3.8 Live and Live Extended Thinking?

Multimodal AI models work with several kinds of data at once, such as text, images, audio and video, instead of text alone. A hard part of voice assistants is real-time conversation: they must respond with low delay, take in what the user shows them, and let people interrupt. Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking are Google models for this job, taking audio, images, video and text as input and giving audio and text as output. Separately, Google's Live API documentation describes a Live API for Gemini that processes continuous streams of audio, images and text to give immediate spoken responses, and the maker describes Live Extended Thinking as reasoning and speaking at the same time, for tasks needing deeper reasoning. Google DeepMind published their model card in September 2026.

Written from primary sources and reference pages, and checked against them by an independent AI fact-checker. Drafted with AI assistance.