OpenAI launches GPT-Live, a new generation of full-duplex voice models that make ChatGPT conversations far more natural. Still limited to the ChatGPT interface, its future arrival in the API will be a real turning point for professional voice agents.
Published on July 9, 2026 | AI Voicebots & Voice Agents
OpenAI has just launched GPT-Live, a new generation of voice models capable of listening and speaking at the same time, making conversations with ChatGPT noticeably more natural. For now, these models are only accessible inside the ChatGPT interface; their future arrival in the API would represent a real turning point for everyone building professional voice agents — including Versatik.
1. GPT-Live: A New Generation of Voice AI
OpenAI announced GPT-Live as a family of voice models designed to make ChatGPT conversations far more fluid — closer to a human exchange than a classic human-machine dialogue. Concretely, GPT-Live introduces a full-duplex architecture: the AI can listen and speak at the same time, punctuate the conversation with small listening cues ("mhmm", acknowledgments), handle interruptions, and pick the thread back up where the user left it.
One of the key goals of GPT-Live is to eliminate the rigidity of the "turn-by-turn" mode that has characterised most voice assistants until now. Instead of waiting for you to finish speaking before responding, the AI behaves like a conversation partner that reacts continuously, adapting to your pace, silences, topic changes and digressions.
2. Two Models for ChatGPT: GPT-Live-1 and GPT-Live-1 Mini
GPT-Live isn't a single model, but a small family:
- GPT-Live-1, the flagship model, designed for rich, versatile, demanding conversations. At launch, it relies on GPT-5.5 in the background for tasks requiring web search, deeper reasoning, or complex actions.
- GPT-Live-1 mini, a lighter version, optimised for consumer use cases and more constrained infrastructure.
OpenAI has started rolling out these models directly inside ChatGPT Voice, globally:
- Paid users get GPT-Live-1 as the new default voice model.
- Free users are progressively switched to GPT-Live-1 mini, the new voice standard for the free tier.
Worth noting: at launch, GPT-Live does not yet support voice combined with video or screen sharing in ChatGPT — a limitation OpenAI plans to lift soon.
3. Why GPT-Live Is a Turning Point
For Versatik, and for the voice agent ecosystem in general, GPT-Live represents a clear break on several levels.
A New User Experience Standard
GPT-Live raises the bar for what's expected from a voice interface: the ability to listen and speak simultaneously, natural handling of interruptions and topic changes without breaking the flow, and continuous backchannel signals showing the AI "is there" even when it isn't speaking at length. This fluidity is becoming the de facto new standard on the ChatGPT side — every consumer or professional voice product will now be compared to this experience.
A Deep Impact on Existing Architectures
Until now, most voice agents relied on "half-duplex" flows: send an audio request, receive text, then possibly an audio response. With GPT-Live, the AI must handle a continuous incoming and outgoing audio stream, which requires different client-side orchestration (WebRTC, fine-grained microphone handling, audio mixing). Current stacks built on API voice models (GPT-Realtime, Whisper, Translate...) will need to evolve toward this full-duplex behaviour.
4. GPT-Live Isn't in the API Yet: A Crucial Point
Let's be very clear: as of this writing, GPT-Live-1 and GPT-Live-1 mini are not available via the OpenAI API. Both models are exclusively accessible inside the ChatGPT interface (web, iOS, Android) via Voice mode.
OpenAI offers a pre-registration form for "GPT-Live-1 in the API," letting businesses and developers register to be notified as soon as the model is exposed on the platform. We're therefore in a transition phase: the general public is discovering the new voice experience in ChatGPT, while the integrator community waits for the API rollout to be able to replicate it in their own products.
Meanwhile, the voice agents being built today rely on the models already available in the API: the GPT-Realtime family for real-time audio conversations, along with the transcription and voice generation models used for months across B2B use cases (support, sales, education, media...).
5. GPT-Live vs Current API Voice Models: What Changes
On the ChatGPT interface side, GPT-Live delivers a "human-like" voice experience, intelligent routing to the best available text models for reasoning, and immediate global availability. The logic is UX-centred: OpenAI is redefining what it means to "talk to an AI" for millions of users, without them having to manage any technical complexity.
On the API side, the voice models already available remain the backbone of professional architectures today: GPT-Realtime is designed for real-time interactions where the developer controls the flows, latency, behaviour and integration with their systems (CRM, ERP, business databases). These models already provide a very solid base for building voice agents — the difference with GPT-Live mainly comes from the sophistication of the conversational layer (full-duplex, micro-reactions) that the developer doesn't have to orchestrate manually.
6. Why Versatik Is Closely Following GPT-Live's Arrival in the API
At Versatik, which helps businesses design AI agents for customer support, sales, or B2B relationships, GPT-Live is a development we're watching closely.
Organisations deploying voice agents are no longer just looking to "answer correctly": they want natural, fast interactions capable of handling complex situations without breaking the conversation. GPT-Live-1's arrival in the API would mean the ability to offer clients voice agents that speak and listen at the same time like ChatGPT Voice, with the same conversation quality as the consumer interface — but in demanding B2B contexts (export, sales, service, training).
While waiting for this rollout, Versatik continues to leverage the Realtime and voice models already available in the API for concrete use cases, while designing modular architectures able to migrate quickly to GPT-Live-1 as soon as access opens up.
7. Conclusion: A Strategic Wait
With GPT-Live, OpenAI is putting voice at the centre of the AI experience and setting a new standard of naturalness for conversations. For now, this revolution is playing out inside the ChatGPT interface, where GPT-Live-1 and GPT-Live-1 mini are replacing the older voice models.
For everyone building voice agents on the API — including Versatik and its clients — GPT-Live-1's rollout to the platform will represent a real turning point: a new way of designing voice, the user relationship, and the technical architecture. While waiting for that availability, Versatik continues to capitalise on existing API voice models, while preparing to integrate GPT-Live as soon as the API is ready.
> "Want a voice AI agent built on the most advanced technologies on the market, ready to evolve with the next generations of models? Contact us for a POC or a personalised demo."
Learn more: AI voice agents · AI agents — or write to us at contact@versatik.net.