In the course of the deep integration of 5G and artificial intelligence, challenges have arisen for artificial intelligence networks such as diversified scenarios, dispersed applications and limited model generalizability. Addressing these issues requires gaining knowledge from massive heterogeneous network data to summarize the capabilities for different scenarios and services, ultimately forming a generalizable large model as a ubiquitous social service model. Moreover, mobile networks must serve as the foundation for pervasive intelligence, providing continuous connectivity and efficient on-demand artificial intelligence services. In this article, we propose a converged large model and network approach to address the above challenges. We introduce the basic concepts and components of large models, clarify their technical characteristics, and propose an end - to-end convergence architecture. Finally, we identify the key technical issues that require further investigation.