Google has officially launched Gemini 3.8 Live, a new real-time voice model, alongside a specialized version called Gemini 3.8 Live Extended Thinking. These models utilize a native audio-to-audio architecture designed to enhance natural interaction and reduce latency in voice-based AI communication.
The Extended Thinking variant introduces a 'Think as it speaks' capability. This allows the system to perform complex multi-step logic, code debugging, or technical diagnostics in the background while maintaining a continuous, real-time conversation without forcing the user to wait for a response.
The new models are designed for high-complexity tasks, including real-time troubleshooting, mathematical reasoning, and simultaneous language tutoring. Google reports that these models have achieved top rankings in various multimodal and voice reasoning industry benchmarks.
Developers and enterprise users can access these capabilities starting today via the Gemini API and Google AI Studio. The low-latency audio API is intended for integration into smart hardware, customer service platforms, programming assistants, and collaborative office software.
