Google DeepMind just dropped two new audio models, Gemini 3.8 Live and 3.8 Live Extended Thinking, aiming directly at OpenAI’s dominance in real-time voice interaction. Available now via the Gemini API and Google AI Studio, these tools let developers build voice agents that handle background API calls and process visual data without dropping the conversation. The Extended Thinking variant isn’t just talking a big game; it currently sits at the top of the Artificial Analysis Speech-to-Speech Leaderboard with an 82.6 percent score, edging out OpenAI’s GPT-Live-1.
The real story here is the price tag. Google is charging $0.005 per minute for audio input and $0.018 for output. Compare that to OpenAI’s $0.05 per minute rate for GPT-Live-1, and the difference becomes stark. An hour of continuous voice conversation runs about $1.38 on Google’s infrastructure, whereas you’d pay at least $3.00 with OpenAI. For startups building customer service bots or interactive companions, that 54 percent cost reduction could be the difference between a viable business model and burning cash.
But cheaper doesn’t always mean better. OpenAI’s model still holds the edge in conversational fluidity thanks to full duplex technology, which allows simultaneous listening and speaking. Early demos suggest Google’s audio quality lags slightly behind, indicating a strategic choice to optimize for affordability and raw benchmark performance rather than premium user experience. It supports over 97 languages, which is a massive win for global deployment, but the nuance in tone might feel robotic compared to its rival.
This move fits a familiar pattern for Google: leverage scale to drive down costs and force competitors to react. While OpenAI focuses on polishing the human-like quality of its interactions, Google is betting that developers will prioritize margin-friendly infrastructure. Sample apps are already live on GitHub for those wanting to test the waters. The question now is whether enterprises will tolerate a slight dip in naturalness for significant savings, or if voice AI remains a space where quality trumps price.
