AgiBot Says Humanoid Robots Could Reach Their ‘GPT-3.5 Moment’ Within Five Years
Humanoid robots could reach a breakthrough in general-purpose capability within the next three to five years, according to AgiBot co-founder Yao Maoqing, as Chinese robotics companies shift their...
Humanoid robots could reach a breakthrough in general-purpose capability within the next three to five years, according to AgiBot co-founder Yao Maoqing, as Chinese robotics companies shift their focus from demonstrations towards reliable deployment in factories and service environments.
Quick Skim
Speaking at the Fortune Leaders Forum in Macau, Yao compared the potential turning point for embodied AI with the arrival of GPT-3.5, when generative AI became capable enough to handle a much wider range of everyday tasks.
AgiBot Targets Broader Robot Intelligence
Yao expects continued improvements in embodied AI models to eventually allow robots to perform a wider variety of physical tasks rather than being programmed for narrow applications.
He said many robotics developers expect embodied AI to benefit from a similar scaling effect to large language models: as training data and model capabilities increase, robot intelligence could improve significantly.
The challenge is turning that intelligence into machines that can operate consistently in unpredictable physical environments.
Factory Reliability Remains the Main Barrier
For industrial users, technological sophistication alone is not enough.
Yao said manufacturers primarily evaluate robots according to four practical measures: task success rate, cycle time, stability and cost.
This creates a much higher reliability requirement for factory robots than for systems used mainly in demonstrations.
A robot performing the same manufacturing process thousands of times must complete the task consistently without disrupting production, making reliability one of the biggest obstacles to widespread humanoid adoption.
AgiBot Reports 99.99% Task Success Rate
AgiBot has been working to demonstrate that its robots can operate consistently across repetitive manufacturing tasks.
In June, the company conducted a six-day livestream involving more than 64,000 manufacturing tasks, reporting a 99.99% completion success rate.
Yao said achieving that level of performance required intensive training sessions over the previous month.
The demonstration reflects a broader shift across the humanoid industry towards proving performance through sustained industrial tasks rather than short demonstrations.
AgiBot Leads Humanoid Shipments
AgiBot has also emerged as one of China’s largest humanoid robotics manufacturers.
The company shipped approximately 9,700 humanoid robots during the first half of 2026, ranking it first globally according to Counterpoint Research data cited by Fortune.
The Shanghai-based company is also considering an initial public offering in Hong Kong as it expands production and commercial deployments.
The scale of shipments suggests China’s humanoid robotics industry is moving increasingly quickly from prototype development towards larger-volume manufacturing.
Hotels and Factories Become Early Deployment Areas
Other Chinese robotics companies are similarly searching for applications where humanoids and service robots can deliver measurable operational value.
Shanghai-based Keenon Robotics, for example, has deployed five robots at a hotel to perform tasks including guest reception, room-service delivery, cleaning and restaurant operations.
UBTech, meanwhile, is concentrating heavily on industrial applications while initially targeting tasks with sufficiently large markets to justify development costs.
Factories, logistics operations and commercial-service environments are therefore emerging as early testing grounds for general-purpose robotics.
China Pushes Humanoids Beyond Demonstrations
China’s humanoid robotics sector has attracted attention through running competitions, boxing demonstrations and other public showcases.
However, developers increasingly see commercial deployment as the more important benchmark.
For manufacturers, humanoid robots will need to demonstrate that they can operate alongside existing automation while delivering predictable productivity and cost benefits.
AgiBot’s focus on reliability shows how the industry is moving from asking what humanoid robots can demonstrate towards whether they can perform useful work repeatedly under real production conditions.
If Yao’s three-to-five-year timeline proves accurate, the next major breakthrough in Physical AI may be defined less by increasingly impressive demonstrations and more by robots becoming reliable enough to work across factories and everyday commercial environments.
