Google Launches Gemini 3.8 Live AI with Advanced Voice Interaction
New models enable real-time reasoning and multitasking during voice conversations.
1Next-gen voice AI capabilities
Gemini 3.8 Live introduces near real-time visual processing and 97-language automatic detection during conversations. The Extended Thinking version simultaneously reasons and speaks, using verbal cues like 'Let me check that' to maintain natural flow. Both models execute background tasks without interrupting dialogue, handling complex workflows from chess games to business plan creation through voice alone.
2Enterprise-grade performance benchmarks
3.8 Live Extended Thinking leads the Speech to Speech Quality Index with 82.6 points and achieves 68.6% on agentic task completion benchmarks. It scores 97.7% on Big Bench Audio for reasoning while maintaining competitive pricing. The standard 3.8 Live model ranks second in Speech Agent Arena preference tests, offering cost-effective scaling for developers.
3Developer integration ecosystem
The Gemini Live API enables platforms like LangChain and Vercel to build voice interfaces easily. Partners including Salesforce and Lumeris highlight the model's low latency and tool-calling capabilities. The system handles real-time media streaming infrastructure, letting developers focus on UX design for voice-driven applications.
4Multimodal workplace applications
Integrated across Google Workspace, these models transform Docs, Gmail and Keep with live collaboration features. They process visual context for real-time employee onboarding and convert sketches into React components through voice feedback. The AI coordinates multi-step bookings and asynchronous functions while maintaining conversation flow.
5Technical safety measures
All AI-generated audio includes SynthID watermarks for content detection. The imperceptible digital signature helps combat misinformation while preserving audio quality. Google provides detailed model cards outlining safety protocols and responsible deployment practices for enterprise implementations.
6Real-time language processing
The models automatically transition between supported languages mid-conversation, enabling natural multilingual dialogues. Background API execution allows continuous chatting during task processing. This creates fluid interactions where the AI acknowledges requests immediately while working on solutions behind the scenes.
7Agent workflow optimization
On ServiceNow's EVA-Bench, Gemini 3.8 redefines the Pareto Frontier for voice agents by balancing accuracy with conversational quality. The Extended Thinking version narrates progress during multi-step tasks, guiding users through complex processes like financial analysis or marketing toolkit creation via natural speech.
8Conversational AI evolution
These launches represent a paradigm shift from command-based to reasoning-capable voice assistants. By handling interruptions and explaining thought processes, the models achieve human-like dialogue fluidity. The technology sets new standards for AI collaboration in both consumer and enterprise environments.
Original article: DeepMind — Google Launches Gemini 3.8 Live AI with Advanced Voice Interaction