Based on arXiv:1705.07935 and arXiv:1705.07917, we present the first global fits of supersymmetric models using the new Global And Modular Beyond-the-Standard-Model Inference Tool (GAMBIT). With GAMBIT we have performed frequentist fits of the GUT-scale CMSSM, NUHM1 and NUHM2 models, as well as the weak-scale MSSM7, extending existing results in terms of the number of observables included, scanning techniques and treatment of nuisance parameters. For the GUT-scale models our analyses show that a stop co-annihilation scenario provides the best fit to current data, and that in the CMSSM the stau co-annihilation scenario is ruled out at 95% confidence level. For the MSSM7 we find that the best-fit scenario has light higgsinos and highly under-abundant relic density due to efficient chargino co-annihilation.