Google has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two voice processing models designed to reduce latency in AI agents. These models enable near-real-time reasoning and allow for simultaneous speech and thought processing, alongside the ability to execute third-party software tool calls in the background.
The Extended Thinking variant achieved a score of 82.6 on the Artificial Analysis Speech to Speech Quality Index, outperforming GPT-Live-1-Astra and Grok Voice Think Fast 2.0. Additionally, the models recorded scores of 68.6% on T-Voice, 35.1% on T-Voice-banking, and 97.7% on Big Bench Audio.
The new architecture allows AI agents to perform API calls while maintaining a continuous conversation. Features include automatic language detection with support for 97 languages, the ability to switch languages mid-conversation, and near-real-time visual grounding.
Pricing for the standard Gemini 3.8 Live model is set at $0.005 per minute for audio inputs and $0.018 per minute for outputs. The Extended Thinking model includes additional charges for reasoning tokens and processing inputs such as video and documents.
