Gemini 3.8 Live Extended Thinking: Google's Reasoning Voice Model

What Gemini 3.8 Live Extended Thinking is

Gemini 3.8 Live Extended Thinking is a Google audio-to-audio model announced in Google's September 2026 updates alongside Gemini 3.8 Live. Google's model documentation describes it as a "high-reasoning audio-to-audio model recommended when higher background reasoning is required", for complex, multi-step problem solving during real-time voice interactions. Google's update post calls it the top-rated speech-to-speech model for multi-step reasoning; that ranking is Google's claim. The model code is gemini-3.8-live-extended-thinking, listed as stable.

How it works

According to Google's model page, the model processes background reasoning and asynchronous tool calls while it streams continuous audio responses. That is the difference Google draws with Gemini 3.8 Live, which its model index describes as the default model for low-latency voice agents "without reasoning delays". Function calling on the Extended Thinking model is listed as asynchronous only.

Inputs, outputs and limits

The model accepts text, images, audio and video, and returns text and audio. The input token limit is 131,072 and the output limit is 65,536 tokens. Google lists audio generation, function calling, the Live API, search grounding and thinking as supported. Caching, code execution, file search, grounding with Google Maps, image generation, structured outputs, URL context and the Batch API are listed as not supported. The page gives September 2026 as the latest update and does not state a knowledge cutoff.

Pricing

Google's pricing page lists one price table for Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking and Gemini 3.1 Flash Live Preview, so the Extended Thinking model costs the same per token as the standard model. On the paid tier, text input is $0.75 per million tokens and text output $4.50, with output prices including thinking tokens. Audio input is $3.00 per million tokens or $0.005 per minute, audio output $12.00 per million tokens or $0.018 per minute, and image or video input $1.00 per million tokens or $0.002 per minute. Because thinking tokens are billed as output, a response with more background reasoning produces more billed output tokens.

Status in Google's model index

Google's Gemini API model index lists both Gemini 3.8 Live models as new stable models, and describes the Extended Thinking one as recommended when higher background reasoning is required. The shared price table also covers Gemini 3.1 Flash Live Preview, which remains listed as a preview. Unlike Gemini 3.8 Live, the Extended Thinking model lists the Batch API as not supported, and its function calling runs asynchronously while audio continues to stream.

Where it is available

Google states both Gemini 3.8 Live models have rolled out for developers in Google AI Studio and the Gemini API, and for everyone across Search Live, Gemini Live and Google Workspace. The free tier on the pricing page lists input and output as free of charge, and search grounding includes 5,000 free requests a month shared across Gemini 3 and newer models before $14 per 1,000 requests.

Official Resources