Google Releases Gemini 3.8 Live Audio Models at Lower Cost Than GPT-Live-1

My take: Google just moved the floor on real-time voice AI models. The new Gemini 3.8 Live is priced at $0.84 per hour of input audio at the standard tier, and $3.50 per hour for Extended Thinking, while OpenAI's equivalent, GPT-Live-1 Astra, costs $5.83 per hour.

That is not a minor adjustment: the Extended Thinking tier costs 40% less than OpenAI's, and the standard model is roughly seven times cheaper. For any company or developer building voice tools or conversational assistants, that cost difference can determine whether a product is viable.

That said, the quality scores Google published were measured by Google itself, which has an obvious incentive to make its own models look good. Those numbers are worth reading critically and waiting for third-party validations before treating them as definitive.

The question for businesses and developers building with voice: are you comparing real costs at scale across providers, or still using the one you adopted first?