Google Deepmind has released Gemini 3.8 Live and 3.8 Live Extended Thinking, two new audio models accessible via the Gemini API and Google AI Studio. These tools enable voice agents to execute background API calls, process visual input, and maintain conversation simultaneously. The Extended Thinking variant currently holds the top position on the Artificial Analysis Speech-to-Speech Leaderboard with a score of 82.6 percent, surpassing OpenAI‘s latest GPT-Live-1 models. Developers can access sample applications on GitHub.
Pricing for the new models stands at $0.005 per minute for audio input and $0.018 for output, which is significantly lower than OpenAI’s GPT-Live-1 rate of $0.05 per minute. An hour of voice conversation costs approximately $1.38 with Google compared to at least $3.00 with OpenAI. While the pricing advantage is clear, OpenAI’s model continues to offer more natural conversations through full duplex capabilities that allow simultaneous listening and speaking. Judging from available demos, the OpenAI output also sounds superior, suggesting Google prioritises price over quality.
* Supports over 97 languages
* Available through the Gemini API and Google AI Studio
* Sample apps hosted on GitHub
