Overview
- Google launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on Sept. 15 and made both models available to developers through the Gemini API and Google AI Studio while rolling them into select Search Live and Workspace experiences.
- Artificial Analysis measured Extended Thinking at 82.6 on its Speech‑to‑Speech Quality Index which narrowly beat OpenAI and xAI but represents a small, single‑benchmark lead.
- Extended Thinking is designed to speak interim responses while running background reasoning and nonblocking tool calls, so interfaces must track interaction_status rather than voice output to know when a task is truly finished.
- Google is offering private enterprise previews and is working with partners including Salesforce, Lumeris, and Genspark to develop production use cases for the new voice models.
- Industry observers warn that leaderboard positions shift quickly so buyers should validate real workloads, latency, cost (Google lists separate audio and extended‑reasoning fees), and integration behavior rather than relying on one benchmark result.