Land use models play an important role in exploring future land change dynamics and are instrumental to support the integration of knowledge in land system science. However, only modest progress has been made in achieving these aims due to insufficient model evaluation and limited representation of the underlying socio-ecological processes. We discuss how land use models can better represent multi-scalar dynamics, human agency and demand-supply relations, and how we can achieve learning from model evaluation. By addressing these issues we outline pathways towards a new generation of land use models that allow not only the assessment of future land cover pattern changes, but also stimulate envisioning future land use by society to support debate on sustainability solutions and help design alternative solutions.