Gemini 3.8 Live

Overview

Gemini 3.8 Live is Google DeepMind's production-grade real-time audio model designed for low-latency, speech-to-speech AI applications including voice agents, multilingual conversations, and tool-calling workflows. It ships with a companion variant, Gemini 3.8 Live Extended Thinking, which adds extended reasoning depth for complex queries without abandoning real-time responsiveness. The model leads public speech-quality benchmarks and is available via API immediately at $0.005 per minute.

Why it matters

Gemini 3.8 Live directly challenges OpenAI's GPT-Live-1 on the two dimensions that matter most for voice-native AI adoption: benchmark performance and pricing. Scoring 82.6 on speech-quality evaluations while undercutting GPT-Live-1 by roughly an order of magnitude at $1.38/hour means Google is applying classic hyperscaler pricing pressure to a nascent but fast-scaling product category. For developers building conversational agents, customer service bots, or real-time translation tools, this pricing gap is not marginal — it changes unit economics materially at scale. The Extended Thinking variant closes the gap between voice and reasoning-heavy use cases, making it competitive for tasks previously requiring a separate orchestration layer. The timing — arriving weeks after Anthropic's reasoning push and amid a weekly cadence of frontier releases — signals that real-time multimodal inference is now a primary battleground in the frontier race.