Digital Twins (DTs) increasingly rely on composed models, i.e. systems assembled from many interacting components, where each component may be represented at a different fidelity and even by a different mathematical model class. A key emerging need is the ability to couple reduced-order models (ROMs) defined on subdomains or components in a way that is stable, accurate, and computationally efficient, while still enabling uncertainty quantification, calibration, and data assimilation. This minisymposium brings together researchers developing the analysis and algorithms for coupling reduced order subdomain models, including domain decomposition-inspired ROM coupling, interface/port methods, projection and constraint-based formulations, and hybrid FOM–ROM couplings. Emphasis will be placed on novel approaches and the mathematical foundations that ensure well-posedness, stability, conservation, and error control across interfaces and across differing fidelities and time scales. Contributions are also welcomed on practical implementations and software support that enable robust, scalable coupling of heterogeneous reduced models in DT workflows.