Gemini 3.8 Live and Extended Thinking: Real-Time Voice and Multi-Step Reasoning

Google launched Gemini 3.8 Live, a real-time voice model supporting 97 languages with automatic language switching, and Gemini 3.8 Live Extended Thinking for enterprise multi-step reasoning. The Thinking version synchronizes real-time thinking with speech, allowing asynchronous tool calls while continuing voice interaction, and near-real-time multimodal vision understanding.
Key insights
Gemini 3.8 Live: High-Concurrency Real-Time Voice
The base Gemini 3.8 Live is designed for large-scale deployment with high concurrency and low latency. It supports real-time voice interaction in 97 languages with automatic language switching and smooth multimodal interaction (vision+voice). It is optimized for everyday assistant and retrieval services with competitive running costs.
Extended Thinking: Real-Time Reasoning with Speech
Gemini 3.8 Live Extended Thinking focuses on enterprise complex tasks with multi-step reasoning and planning. It synchronizes real-time thinking with speech: when faced with a complex problem, it naturally says 'let me check...' while performing background tasks. It supports asynchronous tool calls (API, data retrieval) without interrupting voice conversation. It can also understand live video, such as answering questions about an employee onboarding environment or playing chess.