Gemini 3.8 Live Extended Thinking

Overview

Gemini 3.8 Live Extended Thinking is Google DeepMind's production-grade voice model that combines real-time speech-to-speech inference with an integrated extended reasoning layer. It supports multimodal, multilingual voice interactions and native tool calling, enabling complex, multi-step reasoning within low-latency audio pipelines. The model was released alongside Gemini 3.8 Live as a dual-capability offering targeting voice agent deployments.

Why it matters

Voice-native AI is becoming a primary interface layer for enterprise and consumer applications, and combining real-time audio with deep reasoning in a single model eliminates the latency tax of chaining a speech model to a separate reasoning engine. At $0.005 per minute (roughly $1.38/hour), it undercuts OpenAI's GPT-Live-1 by a significant margin, meaningfully reshaping the unit economics for developers building voice agents at scale. Scoring 82.6 on speech-quality benchmarks at launch while offering production-grade availability signals this is not a research preview — it is a deployable capability. For CTOs evaluating voice infrastructure, this represents a credible alternative to OpenAI's audio stack with a cost structure that can make high-volume voice workloads viable. Investors tracking Google's AI revenue story should note that aggressive pricing here is a market-share play, not a margin play — the question is whether it converts developer adoption into durable platform lock-in.