OpenAI made GPT-Live-1 generally available via its API on 10 September 2026.
The model is designed for full-duplex voice conversations: it can listen and speak at the same time, handle interruptions and continue the conversation whilst a model or agent in the background carries out reasoning or uses tools.
For a freelancer, a coach, a consultant or a small business, the benefit isn’t simply having a more natural-sounding AI voice. The real change lies elsewhere: The telephone can now become a channel directly linked to a schedule, a customer database, an FAQ or a booking system.
In other words, following on from chatbots, AI is now beginning to play a role in one of the most critical moments of the customer journey: when someone calls because they want an answer straight away.
What OpenAI actually announced
According to OpenAI’s official announcement published on 10 September 2026, GPT-Live-1 is available via the API for building voice applications and business workflows. The model can listen and speak simultaneously, handle pauses and interruptions, and delegate certain tasks to another model or to background tools.
OpenAI also states that developers can use it in telephony scenarios. The official release notes indicate that voice sessions are billed 0.05 dollars per minute, charged by the second, to which may be added the cost of the backend model and the tools used.
- OpenAI — Build more natural voice experiences with GPT-Live-1 in the API
- OpenAI — Product and API Release Notes
The key point is that GPT-Live-1 is not a ‘plug-and-play’ telephone system for just any business. It is a infrastructure for developers. Telephony, business rules, permissions, connections to the calendar or CRM, and transfers to a human agent must be built around this.
Why full-duplex is changing the telephone experience
An older voice assistant often operates in turns: you speak, the system waits, it transcribes, it thinks, it replies. This way of working creates silences and makes interruptions difficult to manage.
Full-duplex allows the system to listen and speak at the same time. In practice, this facilitates natural situations:
- a customer interrupts the agent
- he changes his mind right in the middle of a sentence
- he hesitates or pauses
- it simply says ‘yes’, ‘no’, ‘wait’
- he asks for clarification whilst the officer is still speaking
This technical detail has a very tangible commercial consequence: the more natural the conversation seems, the less the automation becomes an additional source of friction.
For VIFLY, this is exactly where things get interesting. Visibility attracts. Understanding reassures. But just when someone is ready to take action, a poor customer journey can ruin everything. If the call goes to an impersonal answering machine or if no one is available to take it, the opportunity may be lost.
Practical example no. 1: a coach who receives enquiries whilst in meetings
Let’s imagine a professional coach who works all day via video calls. During her sessions, she cannot answer the phone.
Hello, I saw your videos on Instagram. I wanted to know if you are also accompanying entrepreneurs and if you have an availability Thursday after 17 hours.
A voice agent linked to validated information could answer common questions and then consult a schedule. If the infrastructure is properly configured, it could offer two slots and prepare a reservation.
This scenario becomes even more logical if the booking system is already structured. A solution like VIFLY Booking can be used to organize availability and booking route; a voice agent could be a conversational layer over this type of infrastructure tomorrow.
The distinction is important: VIFLY Booking does not offer today an integrated GPT‐Live‐1 agent. This is a scenario of evolution made technically more realistic by the arrival of this new generation of voice APIs.
Practical example no. 2: a consultant who assesses a prospect before calling them back
Second case: a consultant receives several commercial calls per week. Many requests are interesting, but others are too vague or out of target.
A vocal agent could ask a few simple questions:
- What is your main need?
- What is the size of your creator business?
- What's your deadline?
- Do you already have a website or online presence?
- Do you want an audit, accompaniment or a full service?
SME of 12 employees. Digital acquisition. Existing site. Launch planned in 6 weeks. Need: audit + support. To be recalled quickly.
The gain is not to remove the consultant. The gain is avoid drawing attention to the first repetitive qualifying minutes.
For an independent who already centralizes his offers and contact points in a LinkHub VIFLY, this logic goes in the same direction: reduce dispersion, clarify offers, and then facilitate action when intent is strong.
The real change: the voice can delegate to other systems
GPT‐Live‐1 can handle the conversation while a model or agent in the background performs another task. This is probably the most important point of the announcement.
- GPT‐Live‐1 manages conversation and rhythm
- a backend model analyses a more complex demand
- a calendar tool check availability
- a CRM recovers or updates client folder
- a reservation system prepare the action
- a human regains hands if the situation becomes sensitive.
This architecture is more realistic than the idea of a single AI capable of doing everything. In a professional system, each component should have a limited role and permission.
This is also why companies that already structure their data, offers and routes will have an advantage. An AI cannot properly consult a schedule that is never up-to-date, or explain an offer whose conditions are contradictory.
What you shouldn’t think: AI shouldn’t take care of everything
The natural voice can give a strong sense of competence. That is precisely why we must set limits.
An agent may be relevant for schedules, availability, rates, simple administrative questions, commercial qualification, preparation of a reservation and follow-up of a standard request.
On the other hand, certain situations should remain under human control: significant litigation, exceptional reimbursement, sensitive legal or medical advice, complex emotional situation, contractual decision or irreversible action.
The right architecture is therefore not "IA or human". It looks more like: IA to treat friction, human to assume judgment and responsibility.
The VIFLY perspective: automate the action, not the trust
Many companies risk making a mistake with voice agents: seek to replace as many human conversations as possible simply because technology allows.
That is not the right objective.
VIFLY logic is based on a simple path: visibility → understanding → trust → action → opportunity.
A vocal agent can be excellent between understanding and action. He can immediately answer a question, reassure about availability, avoid a prospect leaving the course or facilitate a reservation.
But trust must not be automated at all costs. Our recommendation is clear: use IA voice to reduce simple friction, not to remove all human contacts.
For the self-employed, this logic also joins the role of a structured digital presence. With VIFLY, the goal is not just to show links: it is to better organize what the person needs to understand and what they can do next.
What should you do now if you’re self-employed or an entrepreneur?
1. List the grounds of appeal actually received
For a week, notes the reasons why people call: availability, rates, booking, tracking, quote, support, cancellation, urgent request.
2. Classifies each request in three levels
- Information: can an AI respond safely?
- Action: can an AI prepare or perform the action?
- Decision: do we have to have a human?
3. Structure your offers and your answers
A voice agent is as reliable as the information it accesses. Clarify your rates, terms, schedules, cancellation policies, intervention zones and offers.
If your activity is still scattered between several pages and tools, start by centralizing what needs to be understood. The LinkHub VIFLY can serve as a clear entry point for your offers and contact points.
4. Structure your reservation system
If availability exists only in your head or in messages, no reliable automation can work properly. A solution like VIFLY Booking makes the booking stage more structured and simpler for the prospect.
5. Set limits before automating
Put black on white the actions that the agent could read, prepare or perform. Reserve decisions sensitive to a human.
How much does GPT-Live-1 cost?
OpenAI indicates in its version notes that GPT‐Live‐1 voice sessions are currently costing 0,05 dollar per minute, billed per second. The costs of the backend model and the tools used are charged separately.
It would therefore be misleading to say that a full IA telephone standard automatically costs 0,05 dollar per minute. The final cost will also depend on telephony, the model used in the background, the tools used and the product infrastructure.
This is a good example of an important nuance: the price of the model is not the full service price.
Conclusion: after the chatbot, the telephone is becoming programmable
GPT‐Live‐1 does not mean that all small businesses will replace their phone reception tomorrow.
But the announcement of 10 September 2026 shows a very concrete evolution: voice can now become a natural interface to agents, tools and business actions.
For a creator or independent, the most useful question is not "how to replace a person with an AI?".
Which calls actually require my expertise, and which ones exist only because my customer journey still lacks structure?
If you want to prepare your activity for this evolution, start by clarifying your offers, centralizing your contact points and making booking simple. This is precisely the logic that VIFLY seeks to structure: passing an audience from discovery to action without creating unnecessary friction.
FAQ
What is GPT‐Live‐1?
GPT-Live-1 is a real-time voice model of OpenAI designed for full-duplex conversations. He can listen and speak simultaneously, manage interruptions and delegate certain tasks to a model or agent in the background.
GPT-Live-1 can it be used on the phone?
Yes. OpenAI explicitly mentions telephony among the uses supported in its official announcement.
Can he make an appointment automatically?
Not alone. An application must be built around the API and connect the agent to a calendar or reservation system with appropriate permissions.
How much does GPT-Live cost1 ?
OpenAI indicates 0,05 dollar per minute for the voice session, billed per second. The backend models and tools used are invoiced separately.
Should we be a developer to use today?
To directly integrate GPT‐Live‐1 into a business system, yes. For most small businesses, the technology will probably arrive via software that integrates.