Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking for voice agents

Gemini 3.8 Live is tuned for scale and cost efficiency, pairing conversational intelligence with fluid dialogue and visual grounding, while Gemini 3.8 Live Extended Thinking targets high-complexity tasks with multi-step reasoning. Extended Thinking captured the #1 overall spot on Artificial Analysis' Speech-to-Speech Quality Index (82.6), led agentic task completion with 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark, and scored 97.7% on Big Bench Audio, all at what Google calls a competitive price point. Gemini 3.8 Live placed second in the Speech Agent Arena.

Google positions these as production building blocks for reliable voice agents while also powering more fluid spoken interaction across the Gemini app, Workspace, and Search. The launch is part of a wider DeepMind push disclosed this week: Gemini Robotics 2 now controls whole-body movement across robots including Apptronik's Apollo 2, Gemini reached 1 billion users, and Gemini 4 is in the company's most expensive training run ever.

The release intensifies the voice-agent race against OpenAI's real-time models and xAI's Grok Voice, with the benchmark leadership giving Google a concrete marketing claim in a category that has been hard to differentiate. Skeptics will want to see whether the τ-Voice and arena numbers translate to real deployments — speech-to-speech benchmarks have historically diverged from production reliability — and how the cost claims hold up at scale for enterprise voice workloads.