Recent advances in compact thermal models have led to the emergence of a new concept allowing models to be created at any desired order of accuracy. Traditionally, increasing precision was attained by increasing the number of nodes. This strategy faces many problems; in particular, for the case of multiple heat sources (MCM) and∕or stacked dies, because different operating conditions will lead to different temperature and heat flux profiles that will require different node partitioning in order to be matched. In fact, classical approaches face a difficulty in selecting appropriate node size and position, as well as the inability to provide an a priori estimate of the number of nodes needed. The new concept is based on the use of a flexible profile to account for different possible uses of the model. In particular, it can deal with different patterns of heat generation encountered in MCM and stacked dies, and hence it is truly boundary conditions independent. Moreover, the new approach gives access to the tangential temperature gradient. This valuable information for designers in order to assess reliability cannot be predicted by classical compact model approaches. The concept was presented earlier for a simple rectangular 2D structure with surface heating (2004, 10th THERMINIC Conference, pp. 273–280). In this paper, the concept will be generalized to 3D parallelepiped boxes with both surface and∕or volumetric heating. The second achievement is the possibility to deal with geometries that can be decomposed into boxes.