Unitree CEO: GPT-7 Could Make Robots Finally Useful

Imagine telling a robot to clean your kitchen, and it just does it—even if it's never seen your home before. That's the promise of "embodied intelligence," and the CEO of Unitree, a top robotics company, thinks we're on the verge of a huge leap. At a recent event, Wang Xingxing said that OpenAI's new GPT-6 Astra model is so powerful that a future version, GPT-7, could solve the remaining problems within a year or two. That's a big change from his earlier prediction of a decade.

So what does this mean for you? If robots can understand plain language and handle 80% of tasks in 80% of new environments, they could soon help with housework, caregiving, or dangerous jobs. Companies like Unitree are already testing this: Stanford researchers paired a Unitree robot with GPT-6 Astra, and it cleaned a kitchen it had never seen, even finding a specific object when asked vaguely. That's a glimpse of robots that learn on the fly, without special training.

But there's a catch. GPT-6 Astra is also raising security concerns—experts worry it could be used for hacking if it falls into the wrong hands. And while the technology is advancing fast, we're not there yet. Wang's benchmark of 80% success in 80% of new environments is ambitious. Still, with billions invested in AI, the race is on. If he's right, your next coworker or helper might not be human—and it could happen sooner than you think.