Gemini 3.8 Live makes Google’s voice AI think out loud
Google’s Gemini 3.8 Live makes AI chats feel more natural with near-real-time reasoning, visual context, and smoother voice interactions.
Voice AI is getting less awkward. Google has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two speech-focused AI models designed to make conversations with machines faster and more natural.
The idea is that instead of making users wait while an AI searches, calculates or completes a task, the models can keep the conversation going while working in the background.
Keeping the conversation moving
Gemini 3.8 Live is designed for fast, fluid conversations at scale. It can respond to voice, process visual inputs in near real time and continue a conversation while handling tasks in the background. That changes one of the most noticeable frustrations with voice assistants.
Rather than going silent until a task is complete, the AI can continue responding while it works. The model can also automatically detect languages and switch between 97 languages during a conversation.
That could be particularly useful in multilingual markets such as India, where conversations can move between English and local languages naturally.
When AI starts thinking out loud
Gemini 3.8 Live Extended Thinking takes the idea further. Google says the model is designed for more complex, multi-step tasks and can speak while reasoning. It can use short conversational cues to acknowledge that it is checking something, while continuing to work on the task in the background.
For users, that could make AI agents feel less like black boxes. Instead of suddenly going quiet while searching or calling an external tool, the system can narrate its progress as the interaction continues.
The goal is not necessarily to expose every part of the model's reasoning, but to make the experience feel more responsive and understandable.
From conversations to actual tasks
The models also target developers and businesses building voice agents. Google says Gemini 3.8 Live can make tool and API calls in the background while maintaining a live conversation. The models are available through the Gemini API and Google AI Studio, with enterprise access through Gemini Enterprise private preview.
Potential applications include employee onboarding, troubleshooting in Search Live and tasks across Docs, Gmail and Keep. Google also points to more complex workflows, including bookings and creating functional React components from sketches and live feedback.
At the same time, Google is adding SynthID watermarking to audio generated by its AI. The technology is designed to help identify AI-generated audio as voice systems become increasingly realistic.
The models do not remove questions around accuracy, privacy or overreliance on AI agents.
But they point towards a different kind of voice assistant, one that can listen, work, respond and act without forcing the user to wait for every step.