The experimental programmes planned for the next decade are driving developments in the simulation domain: these include the High Luminosity LHC project (HL-LHC), neutrino experiments (LBNF/DUNE), and studies towards future facilities such as Linear Collider (ILC/CLIC) and Future Circular Collider (FCC). The complex detectors of the future, with different module- or cell-level shapes, finer segmentation, and novel materials and detection techniques, require additional features in geometry tools and bring new demands on physics coverage and accuracy within the constraints of the available computing resources. In order to achieve the desired precision in physics measurements, while avoiding that simulation dominates the systematic uncertainties, more accurate simulations and larger Monte Carlo samples will be needed. Therefore, this sets the challenge to develop more accurate models of physics interactions with affordable computing time. The widely used detector simulation toolkit Geant4 is at the core of simulation in almost every HEP experiment. In this paper, we will discuss the status of Geant4 in the context of detector R&D for present and future facilities. We highlight, in particular, the need to review some of the physics models’ assumptions, approximations and limitations in order to increase precision, and to extend the validity of models up to future circular collider energies of the order of 100 TeV. Examples of recent improvements in electromagnetic models will be presented in detail.