Google has launched two new models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, designed for production-grade voice agents. Both models are currently available via the Gemini Live API and Google AI Studio as hosted services, with no self-hosted options provided.
Gemini 3.8 Live focuses on cost efficiency and conversational intelligence, while the Extended Thinking version is optimized for complex tasks. The latter features multi-step reasoning, allowing the model to narrate its progress while executing tasks. It uses verbal cues to acknowledge prompts during background processing.
The Extended Thinking model achieved a score of 82.6 on the Artificial Analysis Speech to Speech Quality Index. It also recorded a 68.6% success rate on the τ-Voice benchmark and 97.7% on the Big Bench Audio test. Google positions these models as alternatives to traditional cascaded speech pipelines.
Pricing for both models is set at $0.005 per minute for audio input and $0.018 per minute for audio output. This is based on rates of $3 per 1 million input tokens and $12 per 1 million output tokens. Developers can integrate these models through partners including Agora, LangChain, and LiveKit.
