On September 16, 2026, Google DeepMind officially activated Gemini 3.8 Live in its public API with a pricing structure that redefines the real-time voice conversation market. At $0.005 per minute of audio input and $0.018 per minute of audio output — approximately $1.38 per hour of continuous conversation — the model positions itself as the most cost-effective option in the market against OpenAI's GPT-Live-1, which charges roughly double. Immediate availability through the Gemini API, Google Workspace, and the Gemini app turns this launch into a concrete opportunity for small and medium businesses to build sophisticated voice agents without the prohibitive costs of the past.
What Did Google DeepMind Announce with Gemini 3.8 Live?
Gemini 3.8 Live is a real-time voice conversation model designed for fluid, low-latency interactions. Its key features include: (1) Competitive pricing at $0.005/min audio input and $0.018/min audio output, establishing a new market floor; (2) A variant called Extended Thinking that processes external tool calls in the background while the user continues speaking, eliminating the awkward pauses that characterized earlier voice assistants; (3) Immediate availability in the Gemini API for developers, native integration in Google Workspace for enterprises, and Gemini app access for end users. The Extended Thinking variant is particularly groundbreaking: it means an agent can query a customer database, run a web search, or call an external API while keeping the conversation active — something no other voice model does natively today.
"With Gemini 3.8 Live, one hour of voice agent for customer service costs under $1.40. For an SMB in Houston handling 500 daily calls, this represents up to 80% savings compared to previous solutions."
Davarion Group & LabsReal Impact for SMBs
- 0124/7 voice customer service: at $1.38/hour, an SMB can run a voice agent all day for under $34/month, answering questions, taking orders, and escalating complex cases to humans
- 02Phone sales agents with real-time inventory access: thanks to Extended Thinking, the agent can check stock, pricing, and availability while the customer is speaking — no 'hold please' moments
- 03Risk to evaluate — regional Spanish accent quality: although Gemini 3.8 is multilingual, businesses should validate comprehension of regional accents (Mexican, Central American) in their specific use cases before going live in production
- 04Immediate action: request access to the Gemini API at ai.google.dev and test the voice API with an internal use case this week — entry cost is low enough to experiment without major commitment
The launch of Gemini 3.8 Live fundamentally changes the economics of voice automation for businesses. Until now, voice AI solutions were the exclusive domain of large enterprises with robust telecom budgets. With costs below $2/hour and a world-class API available today, any business with a developer — or a technology partner — can build a virtual receptionist, collections bot, booking assistant, or automated technical support system. The Extended Thinking feature elevates these applications to a qualitatively different level: the agent doesn't just speak, it acts, queries, and decides in real time while the conversation flows. This opens the door to automations that previously required high-cost human operators.
At Davarion Group & Labs, we have been integrating voice AI APIs for clients in Houston TX and across Latin America for months. With Gemini 3.8 Live, we can now build fully functional voice agents — with access to CRM, inventory, calendars, and ticketing systems — at a fraction of the previous cost. If your business handles customer, prospect, or technical support calls, now is the time to automate. Contact us at davarion.com for a free evaluation of your voice AI use case.
