This paper presents a new experimental and Large Eddy Simulation (LES) database to study upscaling effects in vented gas explosions. The propagation of premixed flames in three setups of increasing size is investigated experimentally and numerically. The baseline model is the well-known laboratory-scale combustion chamber from Sydney (Kent et al., 2005; Masri et al., 2012); two exact replicas at scales 6 and 24.4 were set up by GexCon (Bergen, Norway). The volume ratio of the three setups varies from 1 to more than 10,000, a variation unseen in previous experiments, allowing the exploration of a large range of Reynolds and Damköhler numbers. LES of gaseous fully premixed flames have been performed on the three configurations, under different operating conditions, varying the number of obstacles in the chamber, their position and the type of fuel (hydrogen, propane and methane). Particular attention is paid to the influence of the turbulent combustion model on the results (overpressure, flame front speed) comparing two different algebraic sub-grid scale models, the closures of Colin et al. (2000) and Charlette et al. (2002), used in conjunction with a thickened flame approach. Mesh dependency is checked by performing a highly resolved LES on the small-scale case. For a given scale and with a fixed model constant, LES results agree with experimental results, for all geometric arrangement of the obstacles and all fuels. However, when switching from small-scale cases to medium-scale or large-scale cases this conclusion does not hold, illustrating one of the main deficiencies of these algebraic models, namely the need for an a priori fitting of the model parameters. Although this database was initially designed for safety studies, it is also a difficult test for turbulent combustion models.