The country's first foundational large model in the field of national planning and resources, "Yunyu Sky Model (Professional Edition)", was officially released. This "AI urban planner" with 600 billion parameters integrates remote sensing images, 3D real scenes, planning drawings, and government documents through an innovative architecture of "1 industry foundation large model + 6 specialized intelligent agents", filling a gap in the industry and promoting the scientific and intelligent transformation of megacity governance.

The core competitiveness of this model comes from the country's first specialized corpus for planning and resources - "Kunyu Jinglye Corpus". Its training data includes over 900 authoritative planning documents, 100,000 basic questions and answers, and hundreds of expert dialogues, ensuring a high level of professional depth and dynamic update capabilities.

In practical applications, "Yunyu Sky" demonstrates excellent business cycle processing capabilities: it not only includes ten knowledge modules such as the overall plan of Shanghai, providing round-the-clock decision-making consultation, but also completely eliminates complex GIS operations through natural language interaction, achieving second-level aggregation of multi-source layers. At the same time, the system is directly connected to the business database, supporting the automatic generation of visual analysis from natural language queries to "map-number linkage" analysis, assisting managers in making precise decisions.

Compared to general-purpose robots, this model has stronger multimodal parsing capabilities. Relying on the SenseTime Dayu Dayi large model, it can accurately parse key elements such as building outlines, floor area ratio, and green space ratio in drawings, automatically perform compliance reviews, and improve review efficiency by more than 50%. Based on chain-of-thought technology, the model can seamlessly generate technical reports that comply with national compilation methods.

Currently, "Yunyu Sky" has been put into trial operation in multiple district-level projects in Shanghai. Its autonomous task scheduling engine can complete the entire closed-loop process from identifying illegal construction to retrieving property rights, generating suggestions, and pushing law enforcement, reducing the response speed of planning consultation from "day-level" to "minute-level" and achieving a threefold leap in drawing review efficiency.