Unitree CEO Ties Embodied AI Breakthrough to GPT-7-Class Model
At a Shanghai industry event, Wang Xingxing said a GPT-7-class model could solve embodied intelligence within one to two years, moving up his earlier 10-year forecast.
Unitree CEO Wang Xingxing said at an industry event in Shanghai, China, that a GPT-7-class model could be released next year and may directly solve the problem of embodied intelligence, according to Digital Today. The remarks move up his earlier forecast of how soon the robotics industry would reach its "ChatGPT moment."
Wang also set a maturity criterion for the field: a robot should complete 80% of given tasks in 80% of unfamiliar environments using only voice or text commands. He tied that benchmark to models in the GPT-7 class rather than to today's systems.
In August 2026, Wang predicted it could take 10 years for the robot industry's "ChatGPT moment" to arrive. He said at the Shanghai event that OpenAI's latest model, GPT-6 Astra, has raised expectations for a technical breakthrough.
Stanford Movement Lab combined GPT-6 Astra with a Unitree robot to run an experiment tidying a kitchen it had never seen before. The robot tidied the space and retrieved memorized objects based on under-specified requests, without environment-specific training data or additional policy learning, TechRadar reported.
GPT-6 Astra scored 92.7% on ScreenSpot-Pro, compared with 76.9% for GPT-5.6 Sol. Unitree is known for advanced robot hardware but has recently been investing in AI software development.
Cybersecurity experts are concerned that high-performance AI could escape control and be misused to hack other organizations.
Quick answers
What did Wang Xingxing say about GPT-7?
The Unitree CEO said a GPT-7-class model that may be released next year could directly solve the problem of embodied intelligence.
What did the Stanford Movement Lab experiment involve?
The lab combined GPT-6 Astra with a Unitree robot to tidy a kitchen it had never seen before, retrieving memorized objects based on under-specified requests without environment-specific training data or additional policy learning.
What maturity criterion did Wang set for robots?
He said a robot should complete 80% of given tasks in 80% of unfamiliar environments using only voice or text commands.