Gemini 3.8 Live Extended Thinking

Gemini 3.8 Live Extended Thinking is the higher-reasoning of Google's two Gemini 3.8 Live models, rolled out from 15 September 2026 for complex, multi-step work during real-time voice conversations: Google says it reasons and speaks at the same time rather than pausing to think. Like 3.8 Live it takes audio, images, video and text and answers in audio or text; its Gemini API model page (model id gemini-3.8-live-extended-thinking, listed as stable) gives the same 131,072-token input and 65,536-token output limits, and notes that function calling is asynchronous only. It shares 3.8 Live's price row on the Gemini API pricing page: per million tokens, $0.75 text input, $3.00 audio input ($0.005 a minute) and $1.00 image or video input, with $4.50 text output and $12.00 audio output ($0.018 a minute). Google reports that it holds the top overall spot on Artificial Analysis' Speech to Speech Quality Index at 82.6, and scores 68.6% on τ-Voice, 35.1% on Sierra's τ-Voice-banking benchmark and 97.7% on Big Bench Audio. Those figures are from Google's announcement and are not stored as scores on this board, because no other model here is measured on those benchmarks. Beyond the Gemini API and Google AI Studio, it is rolling out in Gemini Live, in Docs, Gmail and Keep for Google AI subscribers, and in private preview in Gemini Enterprise. Google's model card says it is based on Gemini 3 Pro; its generated audio carries a SynthID watermark.