Google ships Gemini 3.8 Live and 3.8 Live Extended Thinking to the Gemini API with tool calls running in the background during conversation

Sources: primary, discussion
Summary: In a post dated 2026-09-15, Google announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking as available in the Gemini API. Google states that Gemini 3.8 Live executes tools and API calls in the background while continuing the conversation, and that it automatically detects and transitions between 97 supported languages mid-conversation, while Extended Thinking is described separately with asynchronous function calls and live progress narration. The tau-Voice-banking figure of 35.1 percent, tau-Voice at 68.6 percent, Big Bench Audio at 97.7 percent and ServiceNow's EVA-Bench are Google's own runs of third-party benchmarks, while the first place on Artificial Analysis' Speech to Speech Quality Index at 82.6 and the second place in the Speech Agent Arena are placements produced by outside evaluators.
Why it matters: Background tool execution and mid-conversation language switching change how a voice agent is structured, which is a larger developer change than the benchmark table.
Follow-up: Track third-party evaluation of the Live API under real load, in particular background tool execution and mid-conversation language switching.