Gemini 3.8 Live Lets Voice AI Think and Work in the Background

Google has launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new audio-to-audio models designed to make voice agents more capable without making conversations feel like they have stopped. Google says the models are available through the Gemini API, Google Workspace, Search, and the Gemini app.

What Gemini 3.8 Live changes

The main change is what happens when Gemini needs to do work outside the conversation. Gemini 3.8 Live can execute tools and API calls in the background while continuing to talk. That lets the model acknowledge a request, keep the interaction active, and return with the result when the task finishes.

Google also says Gemini 3.8 Live can process visual input in near real time and automatically switch between 97 supported languages during a conversation. Those capabilities give voice agents more context than audio alone, particularly for tasks where users can show the model what they are seeing.

Gemini 3.8 Live Extended Thinking goes further

For harder tasks, Gemini 3.8 Live Extended Thinking adds background reasoning to the live voice session. The model can reason and speak at the same time, using conversational updates such as "Let me check that..." while it works through a multi-step task.

Google's developer documentation describes this as asynchronous reasoning and tool use. A voice session can remain in progress while the model waits for a tool response, then deliver a final answer after the overall task is complete. Developers need to track the session's interaction status rather than treating every turnComplete signal as the end of the interaction.

• Gemini 3.8 Live is designed for low-latency voice agents and direct tasks.
• Gemini 3.8 Live Extended Thinking is designed for complex, multi-step workflows.
• Extended Thinking supports asynchronous function calling while the conversation continues.
• Developers can control Extended Thinking with low, medium, or high thinking levels.

From voice assistant to voice agent

The distinction matters because these models are aimed at tasks that go beyond answering a spoken question. Google shows Gemini 3.8 Live helping with real-time troubleshooting and visual guidance, while Extended Thinking can coordinate multi-step bookings, work from sketches to functional React components, and build business plans through natural speech.

This fits into Google's broader effort to put Gemini across more surfaces. The Gemini app for Windows 10 and 11 gives the assistant a dedicated desktop entry point, while Gemini in Google Sheets on Android brings Gemini-powered spreadsheet analysis to phones.

On mobile, Gemini is also becoming more tightly integrated with the assistant experience. Our guide on how to disable Gemini on Android and switch back to Google Assistant covers the controls Google provides for the default assistant, voice activation, and shortcuts.

The agent direction extends beyond voice. Gemini Spark can now manage Google Photos across multi-step photo tasks, while Gemini Omni 1.1 Flash adds 4K video and 40-second scenes shows Google's push to make Gemini useful across video and other media workflows.

For a broader look at where Gemini fits among today's major AI assistants, see our ChatGPT vs Gemini vs Claude comparison.

What developers need to know

Gemini 3.8 Live is the simpler upgrade path for developers building fast conversational voice agents. Google says asynchronous function calling is now the default for the model, while existing applications can still use blocking tool calls when needed.

Extended Thinking requires a different client-state approach. Because the model can continue reasoning and making tool calls after a conversational response begins, applications should watch interaction_status and treat IDLE as the point when the overall task has finished.

Google lists Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking as generally available in the Gemini API. The models are positioned differently: one emphasizes low-latency conversation and scale, while the other adds deeper reasoning for complex voice workflows.

The practical change is straightforward: voice AI no longer has to choose between sounding conversational and doing meaningful background work. Gemini 3.8 Live is designed to keep the conversation moving while the agent handles tools, reasoning, and other tasks behind the scenes.