OpenAI demonstrates GPT-Live’s multi-task capabilities: chat while checking flights, weather, and stock prices, and hand difficult problems to GPT-5.5

OpenAI released a multi-task demonstration of GPT-Live on July 15, showing that it can maintain continuous conversation while, in the background, looking up flights, weather, stock prices, and itineraries. GPT-Live uses a full-duplex architecture that can both listen and speak at the same time; when it encounters complex tasks, it delegates them to the background GPT-5.5 for processing. On July 8, it officially replaced ChatGPT’s older version of Advanced Voice Mode.
(Background: OpenAI launches GPT-Live: can listen + speak at the same time, creating a human-like conversational experience)
(Additional background: OpenAI’s first hardware has been revealed! A screenless speaker that moves on its own: a built-in camera as your AI companion)

Key takeaways

• On July 15, OpenAI released a multi-task capability demo of GPT-Live, allowing you to look up flights, weather, and stock prices while chatting.
• GPT-Live uses a full-duplex architecture, listening and speaking at the same time; difficult tasks are delegated to the background GPT-5.5, making multiple interactive decisions per second.
• On July 8, GPT-Live replaced Advanced Voice Mode. Paid users get GPT-Live-1, while free users use the mini version.

OpenAI has brought out the tested footage of GPT-Live. On July 15, the company officially released a multi-task capability demonstration, with only one focus: it can keep talking to you without dropping the connection, while handling background chores like looking up flights, checking the weather, grabbing stock quotes, and ranking/organizing your itinerary. It’s only one week since GPT-Live went live.

Compared with the older Advanced Voice Mode’s “you finish speaking and then it speaks” carousel-style turn-taking, this demo is trying to prove not that its voice sounds more natural, but whether it can truly multitask.

Chat while it does things in the background

GPT-Live relies on a full-duplex architecture (full-duplex), allowing it to listen and speak simultaneously. It makes multiple interactive decisions per second—deciding when to speak, when to keep listening, or when to call tools. When tasks require looking up information or multiple steps, it doesn’t freeze the conversation; instead, it throws the work to the background GPT-5.5. Even while computation is ongoing, it continues chatting with you, and visual cards like weather and stock prices can appear in real time during the conversation.

This architecture made its debut when it launched on July 8. GPT-Live-1 is for Go, Plus, and Pro paid users, while GPT-Live-1 mini is for free users—fully taking over Advanced Voice Mode.

The data OpenAI shared at the time was that in human blind tests, both versions clearly outperformed the older version. In the research-level reasoning test GPQA, GPT-Live-1 in its high-reasoning mode scored 84.2%, nearly double the older version’s 45.3%.

Voice should be the primary interface

With this demo added just a week after launch, OpenAI clearly wants to ride the momentum while it’s still hot, driving home the point that “voice is the primary interface, not an add-on to text chat.” That same week, it had just fought a PR battle—earning Codex points based on learnings from GPT-5.6—then followed up with a no-screen voice companion hardware device, consistently going all-in on voice.

The real hurdle has never been whether the voice sounds like a human. It’s whether it can keep talking with you and get things done. What this demo is meant to show is the latter. As for whether it’s actually smooth to use, it’ll only be clear after a large number of users get hands-on.

Common questions

What is GPT-Live’s multi-task capability?

It refers to its ability to, while maintaining continuous conversation, simultaneously handle background tasks like looking up flights, weather, stock prices, and itineraries. Using the full-duplex architecture to listen and speak at the same time, it delegates complex tasks to GPT-5.5 without interrupting the conversation.

How does GPT-Live differ from Advanced Voice Mode?

The older version relies on silent judgment of turn order—once you finish speaking, it switches turns—which can lead to mistakes from pauses and people getting talked over. GPT-Live changes to full-duplex, allowing it to listen and speak at the same time. On GPQA it scored 84.2%, close to nearly double the old version’s 45.3%.