SUMMARY We present a new global model for the Earth’s lithosphere and upper mantle (LithoRef18) obtained through a formal joint inversion of 3-D gravity anomalies, geoid height, satellite-derived gravity gradients and absolute elevation complemented with seismic, thermal and petrological prior information. The model includes crustal thickness, average crustal density, lithospheric thickness, depth-dependent density of the lithospheric mantle, lithospheric geotherms, and average density of the sublithospheric mantle down to 410 km depth with a surface discretization of 2° × 2°. Our results for lithospheric thickness and sublithospheric density structure are in excellent agreement with estimates from recent seismic tomography models. A comparison with higher resolution regional studies in a number of regions around the world indicates that our values of crustal thickness and density are an improvement over a number of previous global crustal models. Given the strong similarity with recent tomography models down to 410 km depth, LithoRef18 can be readily merged with these seismic models to include seismic velocities as part of the reference model. We include several analyses of robustness and reliability of input data, method and results. We also provide easy-to-use codes to interrogate the model and use its predictions for the development of higher-resolution models. Considering the model‘s features and data fitting statistics, LithoRef18 will be useful in a wide range of geophysical and geochemical applications by serving as a reference or initial lithospheric model for (i) higher-resolution gravity, seismological and/or integrated geophysical studies of the lithosphere and upper mantle, (ii) including far-field effects in gravity-based regional studies, (iii) global circulation/convection models that link the lithosphere with the deep Earth, (iv) estimating residual, static and dynamic topography, (v) thermal modelling of sedimentary basins and (vi) studying the links between the lithosphere and the deep Earth, among others. Several avenues for improving the reliability of LithoRef18’s predictions are also discussed. Finally, the inversion methodology presented in this work can be applied in other planets for which potential field data sets are either the only or major constraints to their internal structures (e.g. Moon, Venus, etc.).