<p>The representation of the South American Monsoon System (SAMS) by global climate models (GCMs) is of key relevance for a better comprehension of the physical mechanisms behind the recent and future climate changes over South Tropical South America (STSA) in a global warming scenario. During the last four decades STSA experimented a lengthening of the dry season related to diverse forcings, leading to an increase in fire activity and severe socio-environmental impacts. In the present study, a set of 16 GCMs simulations from the CMIP6 experiment were evaluated during the historical period 1979-2014 in terms of how well they reproduced the atmospheric circulation over STSA through a weather-typing (WTs) approach. 9 WTs were first identified based on low-level wind anomalies from the ERA5 reanalysis, which summarized the atmospheric variations over STSA throughout the year. Focus was put on the representation of WTs during the SAMS initiation and the dry-to-wet transition season (from July to October). Model performance depended on the seasonal cycle and spatial structure of the WTs. Some of the GCMs adequately reproduced the different WTs and their spatio-temporal configurations, with lower skills in the transition seasons. Furthermore, GCMs tended to go from dry to wet conditions too quickly, evidencing deficiencies in the representation of the SAMS onset. This was particularly associated with a poor representation of the southerly wind intrusions to STSA and the intra-seasonal variability of the South American low-level jet. In terms of the relationship between WTs and rainfall on interannual time-scales, a selection of GCMs was able to associate the occurrence of anomalous wet and dry years with specific WTs, indicating well-represented physical processes modulating precipitation variability. Overall, this study could identify few GCMs that managed to simulate the main atmospheric circulation features in STSA (among them, the CESM2, CMCC-CM2-HR4 and MPI-ESM1-2-HR models), which is particularly important for driving high-resolution modelling experiments as well as for the analysis of future projections.</p>