At the Fortune Leaders Forum in Macau on Sept. 8, AGIBOT co-founder Yao Maoqing said embodied AI could reach a 'GPT-3.5' level within 'the 3-to-5-year window.' He said models keep maturing and that 'as data volume rises and your model's parameter count grows, there will definitely be a step-up in intelligence.'
Yao's comparison to GPT-3.5 referred to the OpenAI model behind ChatGPT, which he said was the first time an AI service could perform common, everyday tasks with a success rate between 80% and 90%.
The comments land as AGIBOT tries to turn humanoid robotics into a real business. The company shipped 9,700 units in the first six months of the year, making it the top seller of humanoid robots, according to data from Counterpoint Research released in late August. AGIBOT is also considering an IPO in Hong Kong.
China's robotics sector is moving from spectacle to deployment. Humanoid robots drew attention at the World Humanoid Robot Games in Beijing last August, where they attempted athletic feats including sprinting, weightlifting, and kickboxing. But the companies behind them are now looking for ordinary uses that can justify capital spending.
Keenon Robotics, based in Shanghai, has rolled out robots for hotel services. Wan Bin, the company's chief operating officer, said one Shanghai hotel is using five robots to greet guests, deliver room service, clean rooms, and manage the restaurant floor. He said: 'This scene is one that will become increasingly common in the future.'
Other firms are pushing into industrial work. Jianxin Pang, UBTech's vice president and vice dean of research, said the company has invested heavily in robots for the industrial sector, but is starting with 'relatively common scenarios with a big enough market' to help ensure return on investment.
Yao said the industrial test is still hard. He said it is 'genuinely difficult for today's humanoid firms to develop a robot that can run stably inside a factory.' He added that industry players care about four metrics: 'success rate, cycle time, stability and cost.'
AGIBOT says it ran a six-day livestream in late June showing its robots achieving a 99.99% success rate across more than 64,000 manufacturing tasks. Yao said that milestone required eight-hour sessions in the middle of the night for a month.
The market has already voted on the next phase of robotics. The question is no longer whether the demos can draw headlines. It is whether embodied AI can produce stable machines that work on a factory floor, in a hotel, and eventually at a cost that capital can justify.
For investors and manufacturers, that is the real benchmark. The winners will be the firms that can turn attention into productivity and engineering claims into repeatable output. Everyone else will remain in the demo stage, where the spectacle is loud but the margins are still theoretical.