OpenAI Opens Its Full-Duplex GPT-Live-1.1 Voice Model to Developers at 5 Cents a Minute, as Amazon Polly Lets Brands Clone a Voice From a Few Minutes of Audio

The business phone call just got a lot cheaper to automate. On November 19, OpenAI made GPT-Live-1.1, the voice model that has run voice conversations in ChatGPT Plus and Pro since October 15, available to any developer through its API, according to OpenAI’s developer community announcement and the OpenAI API changelog. The same day, Amazon Web Services announced that voice cloning in Amazon Polly is generally available as a gated release, letting companies build a custom synthetic voice from a few minutes of recordings by a consenting speaker.

Taken together, the two launches put the two hardest pieces of a phone agent, a model that can hold a natural back-and-forth conversation and a voice that sounds like your brand, on public price lists from two of the largest AI providers. For any business that still loses leads to voicemail or long hold queues, that is the real story of the week.

What OpenAI actually released

GPT-Live-1.1 is what OpenAI calls a full-duplex voice model: per its model documentation, it can listen while it speaks, so a caller can interrupt, add a detail or change their mind mid-sentence without the conversation collapsing. Compared with the earlier GPT-Live-1, OpenAI says the new version transcribes speech more accurately and keeps a more consistent accent and voice across languages. The developer announcement adds that it is better at understanding precise details such as letters and numbers, the order numbers, confirmation codes and spelled-out email addresses that trip up most phone bots, and that it follows instructions and completes tasks more reliably.

The key design choice is what OpenAI calls delegation. According to OpenAI’s GPT-Live guide, the voice model handles the conversation itself, while a separate backend model or agent looks things up, calls tools and completes tasks. Developers can let OpenAI run that backend through its Responses API, or connect any model, agent framework or service of their own. The business application still checks permissions, gets required confirmations and runs the functions that touch company systems. In practice that means a caller can keep talking while the backend checks an order status, and the voice agent explains the result when it comes back.

The pricing that matters

OpenAI bills GPT-Live sessions by the minute rather than by the token. The published rates are:

  • $0.05 per minute for an active session.
  • $0.02 per minute in a new “standby” mode, which pauses the model after a period of silence (developers set the idle timeout between 5 and 60 seconds) while the WebRTC call stays open, then resumes when someone speaks.
  • Backend model and tool usage billed separately.

Simple arithmetic shows why this matters. At the active rate, an hour of continuous conversation costs $3 in voice-session fees, and 1,000 five-minute calls come to about $250 before backend reasoning costs. Standby is 60% cheaper per minute than active time, which helps on calls with long pauses, such as a customer hunting for an account number. OpenAI also notes that GPT-Live-1.1 and GPT-Live-1 have separate concurrent-session limits, a detail that matters for any company planning to handle call spikes.

Amazon’s piece: your own brand voice

AWS’s launch post describes Polly voice cloning as a way to give a company one recognizable voice across its interactive voice response (IVR) system, voice agents, mobile app, in-store announcements, advertising and connected devices. Key details from AWS:

  • It launches in seven languages: English, Spanish, German, French, Italian, Portuguese and Hindi. Each cloned voice speaks only the language it was recorded in.
  • Clones run on Polly’s generative engine and are encrypted and access-controlled so only the customer can use them.
  • AWS says it does not keep a copy of the voice recordings once the voice is created and does not use them to train Polly models.
  • Building a cloned voice is free; speech generated with it is billed pay-as-you-go.
  • It is available in 11 AWS Regions, including AWS GovCloud (US-West). Access is already on for select customers; others must apply or go through their AWS account team.

AWS’s own examples lean heavily on customer service: a cloned voice reading out an order number, a delivery window and a confirmation code on a support line, and contact centers updating prompts as offers change without booking a new studio session.

Why it matters for business

Voice is becoming a commodity layer. A year ago, building a phone agent that could handle interruptions and read back an account number correctly took a specialist vendor. Now it is a line item at a few cents a minute. The competitive edge shifts from the voice technology itself to what sits behind it: clean customer data, well-written business rules, tight integration with booking, CRM and payment systems, and clear handoffs to humans.

The architecture is converging. OpenAI’s split between a conversational voice model and a backend that does the real work mirrors how serious voice-agent platforms are already built. That is good news for buyers: it means the “brain” can be swapped or upgraded without rebuilding the phone experience, and companies are less locked into any one model provider.

Brand voice is now a strategic asset, and a risk. Polly’s consent requirement and access controls reflect a real concern: a cloned voice that sounds exactly like a company’s spokesperson is valuable to the business and to fraudsters. Companies adopting cloning should get written consent from voice talent, decide who can generate audio with it, and tell callers when they are talking to AI, which is also the safest course under the patchwork of rules on automated and artificial-voice calls.

The cost math changes staffing decisions. When a minute of AI conversation costs a few cents, the question for a small or mid-size business is no longer whether it can afford to answer every call, but how much revenue it loses each month by not doing so. After-hours calls, overflow during peak times and routine status questions are the obvious first targets.

The bottom line

OpenAI’s GPT-Live-1.1 API release and Amazon Polly’s voice cloning launch, both on November 19, mark the point where natural, interruptible, on-brand phone agents moved from demo to price list. The technology is no longer the bottleneck. Execution is: the businesses that win will be the ones that connect these voices to real systems, guard their brand voice carefully, and make sure no caller ever reaches a dead end.

Put a 24/7 AI Voice Agent on Your Phones Today

You don’t have to build a voice agent from scratch to benefit from this shift. MAJ.COM Voice AI gives your business 24/7 AI voice agents that answer every inbound call and handle outbound sales and customer support, so you never miss a lead. Get started with MAJ.COM Voice AI.

Sources: OpenAI Developer Community, OpenAI API changelog, OpenAI GPT-Live-1.1 model page, OpenAI GPT-Live guide, AWS Machine Learning Blog, AWS What’s New.

Edited by Coron @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous


Discover more from QUE.com

Subscribe to get the latest posts sent to your email.

Coron @QUE.COM

Editor at QUE.com Intelligence.

coronquecom has 206 posts and counting.See all posts by coronquecom

Leave a Reply

Connect with

Discover more from QUE.com

Subscribe now to keep reading and get access to the full archive.

Continue reading