What's this about?
OpenAI showed off improved intelligence in GPT-Live, its real-time voice model: it can now hold a natural conversation while working several tasks at once — checking flights, pulling local weather, and assembling a travel itinerary in parallel, all without dropping the conversational thread.
What exactly is a "voice model"?
Older voice assistants converted your speech to text, ran a text model, then read the answer aloud — slow and stilted. A native voice model listens and speaks directly as audio, so it can be interrupted, react in milliseconds, and talk while it works. The new part here is doing *background work* mid-conversation — closer to a human assistant on the phone who's simultaneously typing.
Why is it worth noting?
No major discussion thread yet (1.1k likes on the demo). The significance: voice + parallel tool use is the form factor labs are betting on for consumer agents, and "keeps talking while doing" is exactly what separates an agent from a chatbot.