Voice AI just got a meaningful price cut and a capability upgrade at the same time. OpenAI released GPT-Live-1 into its API on September 10, 2026, giving developers access to its full-duplex voice model at $0.05 per minute for the voice layer.

This is a big deal for anyone building enterprise voice agents, customer service automation, or internal voice AI tools. Here’s what changed and what it means.

What GPT-Live-1 Actually Does Differently

The headline capability is full-duplex audio: the model can listen and speak at the same time. That sounds like a small technical detail, but it fundamentally changes how voice AI conversations feel. Previous models had to take turns — the AI would stop generating speech to process what the user said. GPT-Live-1 does both simultaneously.

This means it handles interruptions naturally. If a caller changes direction mid-sentence, the agent catches it without awkward pauses or missed intent. It also handles background noise and extended conversations better — two of the most consistent complaints businesses have had when deploying voice AI in real environments.

OpenAI says the model improved by 30 percentage points compared to GPT-Realtime-2.1 on the Full Duplex Bench, its internal benchmark for measuring how well a model handles simultaneous conversation.

How Developers Can Use It

Developers access GPT-Live-1 through the standard API at $0.05 per minute for the voice layer. The model can be paired with backend text models for more complex reasoning or tool use — OpenAI’s own GPT-6 Astra or third-party models both work. This lets you use a fast, cheap voice layer on the front end while reserving more capable (and more expensive) models for the tasks that actually require deep reasoning.

System prompts let developers control tone, speaking speed, and conversation style. So you can build a calm, measured voice agent for a financial services client, and a warm, fast-paced agent for a consumer retail brand, using the same underlying model.

OpenAI Presence: The Enterprise Wrapper

Alongside the API release, OpenAI pointed enterprises to OpenAI Presence, its product that uses GPT-Live-1 to power customer-facing voice agents. The positioning is similar to what you’d expect: agents that answer questions, resolve issues, take approved actions in company systems, and escalate to humans when needed.

This is OpenAI moving further down the stack from model provider to solution provider. It’s no longer just selling API access — it’s competing directly with the category of voice AI companies that have been building enterprise contact center products for the past two years.

What This Means for Business

The cost of voice AI deployments dropped. At $0.05 per minute, a voice agent handling 1,000 calls per month of average 5-minute duration costs $250 in API fees. That’s a number that makes the business case much easier to run for mid-market businesses who found previous pricing hard to justify.

The quality bar moved up. Natural interruption handling and simultaneous listen/speak capability removes one of the key objections businesses have had about voice AI — that it feels robotic and fails in real conversations. GPT-Live-1 narrows the gap between AI voice and human voice significantly.

The market is maturing fast. When OpenAI starts releasing enterprise-grade voice agents and pricing them as commodities, the message for business leaders is that this technology is no longer experimental. Companies that were waiting for voice AI to become reliable enough to deploy are running out of reasons to wait.

For businesses in industries with high call volumes — healthcare, financial services, professional services, retail — this is a capability that your competitors are evaluating right now. The window to be an early mover is closing.

At Enterprise DNA, our Omni Voice service builds voice AI employees that fit your business: handling internal knowledge queries, reporting, team communication, and customer-facing interactions. The underlying models are getting better and cheaper every quarter. If you’ve been thinking about voice AI for your operations, now is a practical time to explore what deployment actually looks like.

Book a call with Sam to discuss your use case