Google on Sept. 15, 2026, introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, its latest native speech-to-speech models for voice agents. Both models opened to developers in the Gemini API and Google AI Studio the same day, with consumer Search Live, Gemini Live, and Workspace Live rollouts starting then.
Gemini 3.8 Live is the scale and cost-efficiency SKU for fluid speech-to-speech dialogue with near real-time visual grounding, mid-conversation switching across 97 languages, and background tool and API calls while the conversation continues. Gemini 3.8 Live Extended Thinking is the high-complexity SKU. Google says it reasons and speaks at the same time, using early verbal cues and live progress narration for multi-step background work.
Company claim: Google says Extended Thinking took the No. 1 spot on Artificial Analysis' Speech to Speech Quality Index at 82.6, led agentic completion with 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark, and scored 97.7% on Big Bench Audio. Google says the base Live model placed second in the Speech Agent Arena. Rollout maps put Live into Search Live and Extended Thinking into Gemini Live, plus Docs for Google AI Pro and Ultra subscribers and Gmail and Keep for Google AI subscribers. Enterprises get both in private preview in Gemini Enterprise.
Independent coverage from 9to5Google and SiliconANGLE reported the same dual-SKU split, Workspace Live surfaces, and company-cited benches on Sept. 15. Artificial Analysis separately published Speech to Speech Index figures matching Google's 82.6 for Extended Thinking (High) and listed standard Gemini 3.8 Live at 76.0. On Artificial Analysis' Speech Agent Arena, Live debuted at Elo 1083, behind only Gemini 3.1 Flash Live at 1096, while Extended Thinking (High) sat lower on preference Elo even as it led the Index.
Gemini API docs list model codes gemini-3.8-live and gemini-3.8-live-extended-thinking, each with 131,072 input and 65,536 output token limits, Live API support, and audio output. Google's developer companion post prices Gemini 3.8 Live audio at $0.005 per minute input and $0.018 per minute output. All product audio carries SynthID watermarking, Google says. What remains open is how far enterprise Customer Experience access expands beyond private preview, and how Extended Thinking's stronger Index scores trade off against lower arena preference versus the base Live model. Company launch tables and third-party Index and Arena scores should be weighed together, not as a single verdict.