Google Launches 'Gemini 3.8 Live' and 'Extended Thinking' with Background Reasoning During Conversations

Google has unveiled two models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, improving real-time reasoning and conversational experiences for voice-based AI agents. This launch focuses on simultaneously handling natural conversation and complex task execution.

Model Features and Performance

Gemini 3.8 Live processes visual input in near real-time to provide context and can automatically detect and switch between 97 languages during conversation. It achieved a balance between accuracy and conversation quality (Pareto Frontier) on ServiceNow EVA-Bench.

Gemini 3.8 Live Extended Thinking is specialized for high-complexity tasks and multi-step reasoning. It ranked 1st overall (82.6 points) on the Artificial Analysis Speech to Speech Quality Index and demonstrated performance of τ-Voice 68.6%, Sierra’s τ-Voice-banking 35.1%, and Big Bench Audio 97.7% on agentic task completion metrics. By performing reasoning and speech simultaneously, it provides initial verbal cues such as "Let me check that…" and narrates background task progress in real time.

Features and Ecosystem Integration

Both models execute tool and API calls in the background while maintaining uninterrupted conversation. This enables converting sketches into React components or supporting complex workflows. Voice-based navigation and real-time drafting are available in Google Workspace's Docs Live, Gmail Live, and Keep Live. Through the Gemini Live API, voice interface development is supported on major platforms including Agora, LangChain, LiveKit, and Vercel. Enterprise partners such as Salesforce and Genspark praised the low latency and tool-calling capabilities.

Accessibility

Enterprise customers can access the models via private preview in Gemini Enterprise, while general users will gain sequential access through the Gemini app and Google AI Pro/Ultra subscriptions in Docs, Gmail, and Keep.