Unitree CEO says GPT-6 humanoids paint and play chess with no task training
At CIIE in Shanghai, Wang Xingxing said real-time frontier-model inference drove brush and pigment sequences; he called the last millimetre of mechanical precision the remaining bottleneck. Claim rests on press reports of his remarks, not a Unitree paper.
Why it matters
Unitree's founder told attendees at China's import expo that humanoids running OpenAI's GPT-6 could paint and play chess without prior task training, including live pigment changes. He framed the hard problem as closing the millimetre-scale mechanical gap, not planning. Unitree's public RobotSkills repository lists GPT-6 skill folders for painting, Gomoku and chess on the G1-D, but no attempt counts or failure rates have been published. Treat capability claims as reported speech until a documented demo lands.
Key facts
- Remarks at CIIE Shanghai on 5 Nov 2026
- CEO Wang Xingxing attributed demos to GPT-6 inference
- Claimed painting and chess with no prior task training
- Cited last-millimetre mechanical gap as bottleneck
- No Unitree technical paper cited in English press