Gemini 3.8 Live and 3.8 Live Extended Thinking

Your voice assistant can now understand interruptions and complex instructions while continuing to chat with you.

Google DeepMind has launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, advanced AI models designed for more natural and intelligent voice interactions. These models excel at complex reasoning, understanding real-time visual context, and executing background tasks without disrupting conversations. Gemini 3.8 Live is optimized for scale and cost efficiency, offering fluid dialogue and visual grounding with support for 97 languages. Gemini 3.8 Live Extended Thinking targets high-complexity tasks with enhanced intelligence and multi-step reasoning capabilities, outperforming existing benchmarks in speech-to-speech quality and agentic task completion. These models are accessible via the Gemini API, integrated into Google Workspace, and available through the Gemini app, aiming to provide developers and enterprises with robust building blocks for voice agents and to enhance user experiences in everyday applications.

The new Gemini 3.8 models represent a significant leap in conversational AI, moving beyond simple command-response interactions. Gemini 3.8 Live's ability to process visual inputs in real-time and seamlessly handle multiple languages mid-conversation signifies a more contextually aware and globally capable AI. Extended Thinking's capacity to manage complex, multi-step tasks in the background, while still maintaining natural dialogue, addresses a key limitation in current voice AI, making it feel more like a true collaborator.

For developers and enterprises, these models offer a powerful foundation for creating sophisticated voice agents that can handle intricate workflows with improved accuracy and conversational quality. The performance gains on benchmarks like the Speech to Speech Quality Index and EVA-Bench indicate that these models are production-ready for demanding applications. The integration into widely used Google products like Workspace and Search also suggests a broader push to embed more advanced, agentic AI capabilities into everyday user experiences.