模型与实验室 4.0 · 优秀 2026-09-15 · 文章

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

Google 发布 Gemini 3.8 Live 与 3.8 Live Extended Thinking 两档实时对话模型:Live 主打流式对话 + 近实时视觉 grounding,97 种语言会话中自动切换,工具调用后台执行不打断对话...

打开原文回到归档

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/ · platform: google-blog · authors: Tom Ouyang, Malini Jaganathan (Gemini Audio Team) · date: 2026-09-15

TL;DR(中文摘要)

Google 发布 Gemini 3.8 Live 与 3.8 Live Extended Thinking 两档实时对话模型:Live 主打流式对话 + 近实时视觉 grounding,97 种语言会话中自动切换,工具调用后台执行不打断对话;ET 档面向高复杂度任务,边推理边说话,拿下 Artificial Analysis Speech to Speech Quality Index 第一(82.6)、tau-Voice agentic 68.6%、tau-Voice-banking 35.1%、Big Bench Audio 97.7%,Live 档居 Speech Agent Arena 第二。两档已通过 Gemini API、Workspace(Docs/Gmail/Keep Live)与 Search Live 提供选型要点:ET High 的 82.6/68.6% 不等于默认 Live 档成绩,实时语音 agent 先改状态机与异步工具再选档。

Summary (English)

Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking: Live targets fluid conversation with near-real-time visual grounding, 97-language mid-conversation switching, and background tool execution; Extended Thinking takes #1 on Artificial Analysis' Speech to Speech Quality Index (82.6), 68.6% on tau-Voice agentic, 35.1% on tau-Voice-banking, 97.7% on Big Bench Audio, while Live ranks second in the Speech Agent Arena. Available via Gemini API, Workspace, and Search Live.

入库依据(同日 digest 交叉验证)

调研卡 2026-09-16-Gemini38Live-选型(价格 $0.005/$0.018 per min 音频、AA/AArena 数据);官方博客全文抓取核实(82.6/68.6%/35.1%/97.7%/97 languages)。