Abstract: The main feature of computers is computational processing, and the value of the Internet lies in sharing and collaboration. Computers give birth to the Internet, and the Internet amplifies the value of computers. The rapid development of the Internet, big data, and cloud computing is pushing artificial intelligence into the era of large models (LMs). At present, the wide application of single LM is facing challenges such as high training cost, low precision and trust, discontinuous service, and the inability to be updated in a distributed manner. However, with the popularity of lightweight, private, and vertical applications, model interconnection is becoming a bottleneck problem that needs to be broken urgently in the development of large models. Drawing inspiration from the development trajectory of the Internet, this paper proposes an open interconnection architecture termed the "AI Model Network" (AI-Modelnet), which aims to establish a self-organizing intelligent network driven by distributed AI models. Through standardized collaboration and routing mechanisms, it enables interconnection, capability sharing, and collaborative reasoning. Firstly, this paper briefly reviews the current research of single LM and multi-model collaboration. Then, the overall framework of AI-Modelnet is articulated systematically and its distinctive value is pointed out. Finally, the development potential of the model internet in large-scale intelligent collaboration was outlined, and key directions for future research were discussed.