Gemini 3.8 Live and 3.8 Live Extended Thinking

Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking are audio models that provide near-real-time reasoning for voice agents. They support 97 languages, visual grounding, and background tool execution. Extended Thinking scores 82.6 on Speech-to-Speech Quality Index, 68.6% on τ-Voice, and 97.7% on Big Bench Audio. Live ranks second in the Speech Agent Arena and balances accuracy with conversational quality on EVA-Bench. Both models are available via the Gemini API, Google AI Studio, and partner platforms, and all generated audio contains an imperceptible SynthID watermark.