GigaAI Founder: World Model Tech Lags Language Models by 3-4 Years
At the 2026 World Model Summit Forum, Huang Guan, founder and CEO of Chinese AI startup GigaAI, spoke during the opening ceremony of the General World Model Beijing Key Laboratory. He said artificial general intelligence should not be limited to language intelligence; it also needs to understand, generate, predict, and act in physical three-dimensional space, which relies on breakthroughs in general world models.
Huang introduced GigaAI's previously proposed five-level system for general world models. L1 can generate visually lifelike general videos and support AI short dramas and game content; L2 can turn environmental understanding into executable action sequences for autonomous driving and embodied robot control; L3 can generate geometrically accurate spatial worlds for locatable, measurable, and navigable 3D environments; L4 aims to construct worlds consistent with physical laws and predict physical cause and effect of actions; L5 targets understanding the physical world in open environments and enabling continuous interaction and dynamic evolution of multi-agent systems.
Huang said world models are relatively mature at L1 and show promise at L2, but have not achieved a breakthrough. He noted they generally lag language models by three to four years, and need action data scaled to at least one million hours for further progress.
The lab, opened by Beijing-based GigaAI, is described as the first laboratory of its kind recognized by a provincial government in China. Lu Jiwen, deputy head of Tsinghua University's automation department and director of the lab's academic committee, said world models predict how things evolve under specific actions. Lin Yonghua, deputy director and chief engineer of the Beijing Academy of AI, said world models have higher computing requirements than language models due to multiple spatial dimensions. Zhu Zheng, co-founder and chief scientist of GigaAI, will serve as director of the lab.
