Physix FrontierPhysix Frontier · News Briefing Card (enbrief · Oct 14, 2026)
Unitree CEO Expects GPT-7 to Crack Embodied AI by 2027
KEY FACTS
- Unitree Technology CEO Wang Xingxing said at an industry event in Shanghai that OpenAI's GPT-6 Astra model is very capable.
- Wang Xingxing believes that GPT-7, to be released next year, may directly solve the core problems of embodied intelligence.
- In August this year, he predicted that this breakthrough would take about ten years, but his expectation has now clearly moved forward.
- He proposed a maturity standard: a robot completes 80% of tasks in 80% of unfamiliar environments using only voice or text commands.
- A Stanford University experiment paired a Unitree robot with GPT-6 Astra to complete a cleaning task in an untrained kitchen.
KEY DATA
92.7%GPT-6 Astra ScreenSp
76.9%GPT-5.6 Sol ScreenSp
80%Task completion rate standard proposed by Wang Xingxing
80%Unfamiliar environment proportion standard proposed by Wang Xingxing
PHYSIX OBSERVATION
Wang Xingxing has compressed the timeline from ten years to one or two, showing that large-model iteration is forcing the robotics industry to reassess its pace. But whether Unitree, strong in hardware and weak in software, can use GPT-7 to close its gaps still depends on the real coupling between model and body, not on benchmark scores. For the industry, the competitive focus of embodied intelligence is shifting from joint motors to model capability.
Source: enbrief original report ↗