Studying large atmospheric molecular clusters is needed to understand the transition between clusters and aerosol particles. In this work, we studied the (SA)n(AM)n clusters with n up to 30 and the (SA)m(AM)m±2 clusters, with m = 6-20. The cluster configurations are sampled using the ABCluster program, and the cluster geometries and thermochemical parameters are calculated using GFN1-xTB. The cluster binding energies are calculated using B97-3c. We find that the addition of sulfuric acid is preferred to the addition of ammonia. The addition free energies were found to have large uncertainties, which could potentially be attributed to errors in the applied level of theory. Based on DLPNO-CCSD(T0)/aug-cc-pVTZ benchmarks of the binding energies of the large (SA)8-9(AM)10 and (SA)10(AM)10-11 clusters, we find that ωB97X-D3BJ with a large basis set is required to yield accurate binding and addition energies. However, based on recalculations of the single-point energy at r2SCAN-3c and ωB97X-D3BJ/6-311++G(3df,3pd), we show that the single-point energy contribution is not the primary source of error. We hypothesize that a larger source of error might be present in the form of insufficient configurational sampling. Finally, we train Δ machine learning model on (SA)n(AM)n clusters with n up to 5 and show that we can predict the binding energies of clusters up to sizes of (SA)30(AM)30 with a binding energy error below 0.6 %. This is an encouraging approach for accurately modeling the binding energies of large acid-base clusters in the future.