AGIBOT's Yao: Robots Reach 'GPT-3.5' Moment in 3–5 Years

The World Humanoid Robot Games in Beijing last August put humanoid robots in the spotlight with athletic feats like sprinting, weightlifting and kickboxing. But behind the headlines, Chinese robot makers are racing to find practical uses in hotels and factories. AGIBOT, which shipped 9,700 humanoid units in the first half of the year and is considering a Hong Kong IPO, is at the center of that push.

At the Fortune Leaders Forum in Macau on Sept. 8, AGIBOT co-founder Yao Maoqing compared where embodied AI stands to GPT-3.5, the model behind ChatGPT. He said the technology will reach that level within a 3-to-5-year window, meaning robots that can perform everyday tasks at an 80–90% success rate. That milestone would mark a shift from specialized demonstrations to general-purpose capability.

But the path to factories is steep. Yao acknowledged that humanoid firms find it genuinely difficult to build a robot that runs stably inside a factory, where industry players care about success rate, cycle time, stability and cost above all else. AGIBOT's six-day livestream in late June showed robots completing over 64,000 manufacturing tasks with a 99.99% success rate, though reaching that required eight-hour overnight sessions for a month. Whether embodied AI delivers on Yao's timeline likely hinges on whether data volume and model size keep producing the same step-up in intelligence seen in large language models — and on whether factories' demands for near-perfect reliability can be met at a cost that makes sense.