Gemini 3.8 Live tops speech-to-speech index

Google DeepMind on Monday introduced Gemini 3.8 Live, positioned for scale and cost efficiency, and Gemini 3.8 Live Extended Thinking, aimed at high-complexity, multi-step tasks. Both began rolling out the same day in the Gemini API and Google AI Studio, and are in private preview in Gemini Enterprise.

On vendor-reported figures, Extended Thinking takes the top overall spot on Artificial Analysis' Speech-to-Speech Quality Index with 82.6, and leads agentic task completion at 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark. It scores 97.7% on Big Bench Audio. Google says the numbers were produced on the Live API running on its Gemini Enterprise Agent Platform. The lighter 3.8 Live placed second in the Speech Agent Arena.

The feature claims are aimed squarely at voice agents: near real-time visual input processing, automatic detection and mid-conversation switching across 97 languages, and background tool and API calls that let the model keep talking while a task completes. Extended Thinking reasons and speaks simultaneously, using verbal fillers such as "Let me check that…" and narrating progress on multi-step background work.

Google named Agora, Fishjam, LiveKit, Pipecat, Vercel and Vision Agents as platform partners, and Salesforce, Genspark and Lumeris as early customers. All generated audio carries SynthID watermarking.

Google did not publish pricing, latency measurements, or parameter details, saying only that the models are competitively priced against other frontier models. Consumer access lands in Search Live for 3.8 Live, and in Gemini Live plus Workspace tiers for Extended Thinking.