OpenAI Launches GPT-Live Voice Models to Power ChatGPT Voice

The News

OpenAI has released GPT-Live, a new generation of voice models that now powers ChatGPT Voice, its biggest overhaul of spoken interaction since Advanced Voice Mode. The company describes the technology as making conversation feel "much more like having a real conversation", built on a full-duplex design that lets the system listen and speak at the same time.

The launch spans a family of models. GPT-Live-1 serves paying Go, Plus and Pro subscribers, while GPT-Live-1 mini becomes the default for free users. Two further variants, GPT-Live-1 Medium and GPT-Live-1 High, draw on GPT-5.5 Thinking with different levels of reasoning effort, and the backend defaults to GPT-5.5 Instant for everyday exchanges.

Unlike older turn-based systems, GPT-Live makes decisions many times per second about whether to speak, pause, listen or interrupt. It can drop in short acknowledgements such as "mhmm", wait while a user thinks, filter background noise and show visual cards for weather, stocks or sports, offloading heavier jobs such as web search to backend models without breaking the flow of talk.

OpenAI says human raters strongly preferred GPT-Live-1 and its mini sibling over Advanced Voice Mode across matched five to ten minute conversations, with gains on the GPQA scientific-reasoning and BrowseComp agentic-search benchmarks. The models are rolling out globally on iOS, Android and ChatGPT.com. An API is not available at launch; developers can register interest through a sign-up form.

Why It Matters

Voice has long been the interface OpenAI struggled to make feel natural. The original Advanced Voice Mode impressed with expressiveness but stumbled over interruptions and the awkward silences that break human rhythm. Full-duplex processing is the harder engineering problem, and shipping it at consumer scale signals a conviction that voice, not text, is the next battleground for assistants.

The timing carries weight. The last time an OpenAI launch reset expectations this sharply was the GPT-4o "omni" demo in May 2024, which set the template every rival has since chased. Google has pushed Gemini Live and Meta has bet on wearables. By folding real-time voice into free tiers, OpenAI wants conversation to become the default way hundreds of millions of people reach ChatGPT, rather than a novelty.

Indian Angle

For India, where much of the next wave of internet users is voice-first and more comfortable speaking than typing, a genuinely conversational assistant matters more than it does in the West. The gap OpenAI flags around non-native accents and fluency in some languages is precisely the opening home-grown players are chasing. Sarvam AI, backed under the government's IndiaAI mission, and Ola's Krutrim have both built speech systems tuned for Indian languages and code-mixed "Hinglish" that global models still handle unevenly.

The missing API is the bigger story for Indian builders. Fintech and support firms such as Razorpay, along with banks and insurers, lean on voice bots for collections, onboarding and service in dozens of languages. Without programmatic access they cannot yet wire GPT-Live into IVR systems or apps, leaving room for local speech providers and keeping costs in rupees rather than dollars.

Regulators will watch too. Real-time voice that filters, records and reasons over conversations sits squarely within the Digital Personal Data Protection Act's consent rules, and the RBI's tightening stance on outsourced customer interactions means firms will need clarity on where voice data is processed before deploying at scale.

FAQ

When does GPT-Live take effect?

It is rolling out now, globally, to existing ChatGPT users on iOS, Android and ChatGPT.com. Free users get GPT-Live-1 mini by default, while Go, Plus and Pro subscribers receive the fuller GPT-Live-1 model. The rollout is gradual, so access may appear over several days.

How is it different from Advanced Voice Mode?

The key change is full-duplex design: GPT-Live listens and speaks at once rather than waiting for you to finish. OpenAI says raters strongly preferred it in matched conversations, citing better handling of pauses, interruptions and background noise.

Can Indian developers use it in their apps?

Not yet. There is no public API at launch, so voice bots and IVR systems cannot integrate GPT-Live directly. OpenAI is collecting interest through a sign-up form and will notify developers when programmatic access opens.

Does it support Indian languages well?

It supports many languages, but OpenAI acknowledges some may carry non-native accents or fluency gaps. For nuanced Indian-language and Hinglish use, locally trained models from Sarvam or Krutrim may still sound more natural today.

This story was reported by OpenAI. Read the full original coverage at OpenAI.