The predictability and forecast skill of the models participating in the Climate Historical Forecast Project (CHFP) database is assessed through evaluating the representation of the upper-tropospheric extratropical circulation in the Southern Hemisphere (SH) in winter and summer and its main modes of variability. In summer, the predictability of 200-hPa geopotential height anomalies mainly comes from the ability of the multimodel ensemble mean (MMEM) to forecast the first three modes of interannual variability with high fidelity. The MMEM can reproduce not only the spatial patterns of these modes but also their temporal evolution. On the other hand, in JJA only the second and fourth modes of variability are predictable by the MMEM. These seasonal differences in the performance of the MMEM seem to be related to the role that the sea surface temperature (SST) anomalies have in influencing the variability of each mode. Accordingly, modes that are strongly linked to tropical SST anomalies are better forecast by the MMEM and show less spread among models. The analysis of both 2-m temperature and precipitation anomalies in the SH associated with the predictable modes reveals that DJF predictable modes are accompanied by significant temperature anomalies. In particular, temperatures at polar (tropical) latitudes are significantly correlated with the first (second) mode. Furthermore, these links obtained with observations are also well forecast by the MMEM and can help to improve seasonal forecast of climate anomalies in those regions with low skill.