Abstract: Starting from the connotation of professional work in urban planning and the typology of artificial intelligence, this paper focuses on the technical paradigm through which large models can be embedded into professional urban planning practice and proposes a pathway for constructing professional planning large models. Urban planning practice follows a diagnosis-inference-treatment working mode and uses domain knowledge to provide solutions. By distinguishing perceptual intelligence from cognitive intelligence, the paper argues that large models can support this working mode. It further proposes a technical framework in which professional planning large models acquire domain knowledge through diagnostic training and treatment training. An empirical case shows that professional planning large models can acquire the planning capabilities of identifying spatial problems, inferring their causes and proposing planning strategies. The findings indicate that the technical paradigm for embedding large models into urban planning practice is based on professional large models equipped with both general knowledge and domain knowledge, and that it supports problem-oriented and law-oriented planning paradigms through the diagnosis-inference-treatment working mode. Professional planning large models are the carrier of this paradigm. Their construction centers on domain knowledge and is realized through diagnostic training and treatment training on the basis of general large models. Domain-specific professional large models can be effectively embedded into planning practice and can enhance planners' professional capacity.