Researchers propose AI-ModelNet, a framework to interconnect and coordinate heterogeneous AI models

A new arXiv preprint introduces AI-ModelNet, a conceptual framework that draws inspiration from the Internet’s architecture to enable interconnection, capability sharing, and collaborative reasoning among heterogeneous AI models. The authors argue that while large models (LMs) have advanced rapidly, practical deployment is constrained by high training costs, deployment complexity, and the proliferation of isolated, domain-specific models. This fragmentation creates a bottleneck in enabling effective interaction and collaboration among models, which the paper frames as a critical unsolved challenge.

The proposed system—AI-ModelNet—is presented as a novel paradigm that establishes pathways between models to facilitate coordination. The paper outlines a systemic vision and hierarchical architecture for AI-ModelNet, detailing how it could support inter-model communication and task delegation. To substantiate the framework’s feasibility, the authors describe a prototype system and present diverse application cases demonstrating its potential utility.

The authors also review the current state of single-model and multi-model research, positioning AI-ModelNet as a next-step evolution beyond isolated model development. They further discuss preliminary directions for future research, emphasizing scalability, standardization, and real-world integration as key priorities for advancing the paradigm.

The preprint is authored by Li Zhetao, Zeng Xiyu, Wang Jianhui, Xiao Yong, Liu Zhongren, Wu Junru, Lai Junjie, Huang Jijun, and Long Saiqin, and was submitted to arXiv on May 25, 2026. It is categorized under cs.AI and includes 31 pages with 14 figures. The work is also slated for publication in the Journal of Computer Research and Development in 2026.