In this work, we will discuss the progress of the QMLMaterial software for efficient global optimization (GO) of atomic cluster and/or spin multiplicity (SM) by means of the active learning (AL) method. The method was applied for the GO of Na20 and 3Al@Si11 atomic clusters. The local optimization of the structures was done using SCC-DFTB and DFT, respectively. For the 3Al@Si11, the doped configurations and the SMs (2, 4, 6) were considered during the search. The Gaussian Process (GP) and Neural Network (NN) algorithms were used with the MBTR structural descriptor in the proposed method.