Google just shipped its most capable live dialogue models yet. Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking landed on September 15, 2026, and the gap between what a voice AI agent can do today versus six months ago is stark.

For businesses evaluating voice AI, this launch matters. Not because Google released another model, but because it signals that the production-grade voice agent stack is maturing fast. The window to get in early is narrowing.

What Google Actually Shipped

Google released two models in this family:

Gemini 3.8 Live is built for scale and cost efficiency. It handles real-time voice dialogue with fluid conversational intelligence, supports audio, image, video, and text inputs, and can execute third-party software tool calls in the background while the conversation is happening. Think customer service agents that check order status, book appointments, or update CRM records mid-call without putting the caller on hold.

Gemini 3.8 Live Extended Thinking is built for complex reasoning tasks. It adds multi-step reasoning with extended thinking time, making it better suited to situations where a voice agent needs to solve a problem rather than just retrieve information. The classic use case: a finance team member asking a voice AI to reconcile a variance in the accounts, not just read out the balance.

Both models support up to 128,000 tokens of context, meaning they can hold significantly more information about a customer, a business process, or a prior conversation than earlier live models.

The Benchmark Story

Performance numbers for the Extended Thinking model are notable. It captured the top position on Artificial Analysis’ Speech-to-Speech Quality Index with a score of 82.6. On the agent task completion benchmark, it hit 68.6% on the standard evaluation and 35.1% on the banking-specific benchmark.

That banking number is worth noting. Financial services, healthcare, and contact centers are the highest-ROI verticals for enterprise voice AI right now, and the domain-specific benchmarks matter more than general leaderboards for businesses in those sectors.

Where It’s Available

The models are accessible today via the Gemini API and Google AI Studio. Enterprise teams can also access them as part of Google Workspace and Gemini Enterprise (currently in private preview for larger deployments).

Google positioned these models as production-ready, not experimental. That framing matters when you’re making a business case internally for deploying voice AI.

What This Means for Business

The speed of improvement in voice AI technology right now has a practical implication: the businesses that start building now will have a significant head start over those that wait.

A few things are shifting simultaneously:

Cost is dropping. Gemini 3.8 Live is priced at less than half the cost of comparable models from six months ago. The economics of running voice AI at scale across customer service, internal operations, or field teams have improved substantially.

Capability is converging. Real-time reasoning, tool execution during calls, and long context windows were enterprise wishlist items eighteen months ago. They’re now available in production models. The gap between “impressive demo” and “we could actually run this in production” has closed.

The interface shift is accelerating. Voice is becoming the primary interface for many AI interactions. A recent survey found 55% of consumers now use voice as their primary AI interface, but only 29% of businesses have deployed customer-facing voice AI. That gap is an opportunity.

For businesses in industries with high call volumes, repeatable workflows, or distributed teams that need to access information quickly, voice AI employees are no longer a future state. They’re a competitive decision being made right now.

The question for most business leaders is not whether to build voice AI capability, but how to approach it without wasting time on the wrong stack, the wrong use cases, or the wrong vendor relationship.

That’s the conversation worth having before the next model update drops in three months.

---

Enterprise DNA’s Omni Voice service deploys voice AI employees for enterprise teams across knowledge discovery, internal reporting, and customer-facing applications. Learn more about Omni Voice or book a discovery call to explore what’s possible for your business.