Gemini 3.8 Live models add background tool calling for voice agents

Google DeepMind introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026, as its most advanced live dialogue models. In DeepMind’s words, they “talk, think, and handle tasks in the background without breaking your flow.”

That background path is the story for anyone shipping voice agents. Standard Live sessions still force a bad trade: answer without planning, or pause while tools finish. Both models support asynchronous tools, but only Extended Thinking is documented as speaking fillers and progress narration while jobs run. Gemini 3.8 Live waits for tool execution before speaking and still allows BLOCKING tools, treating turnComplete as idle. It also processes visual inputs in near real-time and automatically detects and transitions between 97 supported languages mid-conversation. Extended Thinking is confirmed to accept image and video input alongside text and audio.

Gemini 3.8 Live is built for scale and cost efficiency—fluid dialogue with visual grounding. Gemini 3.8 Live Extended Thinking is the high-reasoning audio-to-audio model for complex, multi-step work. It reasons and speaks at the same time, using early verbal cues such as “Let me check that…” and live progress narration while asynchronous tools run.

Extended Thinking is stricter at the protocol layer. Function calling is async-only: declarations must use NON_BLOCKING behavior, and synchronous blocking tools return a hard error. turnComplete: true no longer means the session is idle—the server may still be reasoning or waiting on tools. Clients should watch interaction_status: IN_PROGRESS during processing, IDLE when the session is actually waiting for the user. Thinking depth is low, medium, or high; MINIMAL is not supported. The model accepts text, images, audio, and video, with a 131,072-token input limit and a 65,536-token output limit.

Google reports Extended Thinking first overall on Artificial Analysis’ Speech to Speech Quality Index at 82.6, with 68.6% on τ-Voice and 35.1% on Sierra’s τ-Voice-banking benchmark. Gemini 3.8 Live placed second in the Speech Agent Arena.

Both models are available now in the Gemini API and Google AI Studio. Gemini Enterprise is in private preview. Gemini 3.8 Live is rolling out in Search Live. Extended Thinking is in Gemini Live, in Docs for Google AI Pro and Ultra subscribers, and in Gmail and Keep for all Google AI subscribers. Generated audio is watermarked with SynthID.

Try gemini-3.8-live-extended-thinking in Google AI Studio, then read the Live API thinking guide and switch idle handling from turnComplete to interaction_status before you put a voice agent on it.

Sources