The AI team is delving into several key research directions that could reshape the architecture of next-generation intelligent systems. The first is reasoning ability rather than simple text generation. Frameworks like ROMA and ODS focus on how problems are decomposed, solved in parallel, and then integrated to simulate the complete human thinking process. This multi-step reasoning approach can handle more complex scenarios compared to directly generating answers. The second is multi-agent collaboration models. Unlike the traditional design of a single ultra-large-scale model, the research team is exploring how multiple intelligent agents can work together, with each agent focusing on specific tasks and completing complex workflows through collaborative mechanisms. This distributed approach presents new possibilities in resource efficiency and modular management. Advances in these technological routes will directly impact the development of AI infrastructure in Web3 applications.