Google DeepMind Launches Gemini 3.8 Live and 3.8 Live Extended Thinking Speech Models

Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking near-real-time speech-to-speech models for voice agents; Extended Thinking tops Artificial Analysis' Speech-to-Speech Quality Index at 82.6, with 97-language support, background tool…

Google DeepMind announced Gemini 3.8 Live, a cost-efficient model built for near-real-time dialogue with real-time visual grounding, and Gemini 3.8 Live Extended Thinking, aimed at high-complexity multi-step reasoning in voice agents. Extended Thinking ranks #1 on Artificial Analysis' Speech-to-Speech Quality Index with a score of 82.6, and scores 68.6% on τ-Voice, 35.1% on Sierra's τ-Voice-banking agentic benchmark, and 97.7% on Big Bench Audio. Both models detect and switch among 97 languages mid-conversation, process near-real-time visual input, and execute background tool and API calls, including asynchronous function calling that streams audio while tools execute; MarkTechPost also credits them with alphanumeric precision. Per MarkTechPost, the models are hosted via the Gemini Live API and AI Studio at $0.005/min for audio input and $0.018/min for audio output. Availability spans the Gemini API, AI Studio, Gemini Enterprise private preview, and Search Live; the Hacker News report additionally lists Workspace as part of the rollout. All generated audio carries Google DeepMind's imperceptible SynthID watermark.

• Extended Thinking ranks #1 on Artificial Analysis' Speech-to-Speech Quality Index with a score of 82.6
• Benchmark scores: 68.6% on τ-Voice, 35.1% on Sierra's τ-Voice-banking, 97.7% on Big Bench Audio
• Models detect and switch among 97 languages mid-conversation
• Real-time visual grounding plus background tool and API calls, including asynchronous function calling (MarkTechPost)
• Pricing reported by MarkTechPost: $0.005/min audio input, $0.018/min audio output, hosted via Gemini Live API and AI Studio
• Availability: Gemini API, AI Studio, Gemini Enterprise private preview, and Search Live; Hacker News additionally lists Workspace
• All generated audio is watermarked with Google DeepMind's imperceptible SynthID