Google Launches Gemini 3.8 Live and 3.8 Live Extended Thinking, First on the Artificial Analysis Speech-to-Speech Quality Index

On September 15, 2026 Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, calling them its most advanced live dialogue models yet and rolling them out the same day through the Gemini API, Google Workspace and the Gemini app. The two target different ends of the market: scale and cost efficiency versus high-complexity reasoning that happens while the conversation continues.

How Google splits the two models

• Gemini 3.8 Live: "built for scale and cost efficiency", combining conversational intelligence with fluid dialogue and visual grounding.
• Gemini 3.8 Live Extended Thinking: built for high-complexity tasks with increased intelligence and multi-step reasoning. It reasons and speaks at the same time, using early verbal cues such as "Let me check that..." to acknowledge a request and live progress narration to walk users through multi-step background tasks.
• 3.8 Live processes visual input in near real time, automatically detects and transitions between 97 supported languages mid-conversation, and executes tools and API calls in the background without interrupting the dialogue.

The scores Google lists

The numbers in the announcement cluster around voice-agent benchmarks:

• Extended Thinking takes the top overall spot on Artificial Analysis' Speech to Speech Quality Index at 82.6, leads agentic task completion with 68.6% on tau-Voice and 35.1% on Sierra's tau-Voice-banking benchmark, and scores 97.7% on Big Bench Audio.
• Gemini 3.8 Live secured second place in the Speech Agent Arena, with what Google describes as high user preference.
• On ServiceNow's EVA-Bench, a benchmark for evaluating voice agents, Google says the models push the Pareto frontier for complex workflows by balancing accuracy with conversational quality, noting the run used the Live API on the Gemini Enterprise Agent Platform.

On cost, the post only says Extended Thinking maintains "a highly competitive price point compared to other frontier models" — no per-token or per-minute figures are given.

Where it shows up

Both models began rolling out on September 15: for developers in the Gemini API and Google AI Studio; for enterprises in private preview in Gemini Enterprise, with Gemini Enterprise for Customer Experience coming soon; and for everyone in Search Live (3.8 Live) and Gemini Live (Extended Thinking). On the Workspace side, Docs Live, Gmail Live and Keep Live are named, with Docs for Google AI Pro and Ultra subscribers and Gmail and Keep for all Google AI subscribers.

Google also lists developer platforms building on the Live API — Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel and Vision Agents — plus partners including Salesforce, Genspark and Lumeris.

Provenance and what's missing

All audio generated by our AI products is watermarked with SynthID. — Google's announcement

Google says an imperceptible SynthID watermark is woven into audio output so AI-generated speech stays detectable, and points to the accompanying model card for safety details. What the announcement does not include: concrete pricing, latency figures, or third-party replication of the EVA-Bench result.