Google Launches Gemini 3.8 Live AI Models for Natural Voice Interactions
New AI models enable fluid, intelligent conversations with real-time reasoning capabilities.
1Next-gen conversational AI
Google's Gemini 3.8 Live and 3.8 Live Extended Thinking models represent significant advancements in voice AI technology. These models excel at handling interruptions, switching between 97 languages mid-conversation, and executing background tasks while maintaining dialogue flow. The AI can process visual inputs in near real-time, providing context-aware responses during interactions. Developers can access these capabilities through the Gemini API, Google Workspace, and the Gemini app starting today.
2Enterprise-grade performance
Gemini 3.8 Live Extended Thinking leads the Artificial Analysis' Speech to Speech Quality Index with 82.6 points. It achieves 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark for agentic task completion. The model scores 97.7% on Big Bench Audio for reasoning capabilities while maintaining competitive pricing. These metrics position it as a top choice for production-ready voice agent development, particularly for complex enterprise workflows.
3Developer ecosystem integration
The Gemini Live API enables platforms like Agora, LangChain, and Vercel to build high-performance voice interfaces. Partners including Salesforce and Lumeris highlight the model's low latency and tool-calling capabilities. The API handles real-time media streaming infrastructure, letting developers focus on user experience. This ecosystem support accelerates deployment of voice-driven applications across various industries and use cases.
4Real-time visual context
Gemini 3.8 Live processes visual inputs during conversations, enabling applications like real-time employee onboarding assistance. It can play chess while maintaining natural dialogue flow using visual context. The model transforms raw sketches and voice feedback into functional React components, demonstrating its multimodal capabilities. This visual processing occurs without interrupting the conversational experience.
5Extended Thinking features
The 3.8 Live Extended Thinking model reasons and speaks simultaneously for complex tasks. It provides verbal cues like 'Let me check that...' to maintain natural conversation flow. The model narrates progress during multi-step background tasks, coordinating asynchronous function calls seamlessly. These features make it particularly effective for business planning and custom marketing toolkit generation through voice.
6Workspace integration
The Live models enhance Google Workspace with Docs Live, Gmail Live, and Keep Live functionalities. They enable collaborative document editing and complex task management through natural speech. In Search, the models deliver more intuitive experiences when handling sophisticated queries. The integration focuses on maintaining uninterrupted workflow while executing voice commands.
7Safety measures
All audio generated by Gemini AI products includes SynthID watermarks for content detection. This imperceptible marking helps identify AI-generated content to combat misinformation. Google provides detailed model cards outlining safety and responsibility approaches. The watermarking occurs directly in the audio output without affecting quality.
8Market implications
Gemini 3.8 Live ranks second in the Speech Agent Arena for user preference while remaining cost-effective. On ServiceNow's EVA-Bench, the models advance complex workflow handling by balancing accuracy with conversation quality. These advancements push the boundaries of what's possible in voice agent technology, setting new standards for natural AI interactions in both consumer and enterprise applications.