We tested the library by a large-scale benchmark including a set of 176 non-redundant protein–protein complexes. For each entry in the benchmark, a large number of structural models were generated. Among the derived ensemble of all complex models, we further found the target that has the lowest RMSD from the structure of the native complex. The distribution of the lowest RMSD models for all 176 entries is plotted by the histogram. The figure suggests that the native-like binding is highly likely in the structural ensemble of modeled protein complexes that were built through the library.