AlphaFold 2, which is developed by DeepMind, showed unprecedented success in proteinstructure prediction in the international CASP14 contest, indicating its promisingapplication on protein-peptide structure predictions. Based on the available information, there exists no specific validation conducted on AlphaFold's capability of predicting Gprotein-coupled receptor (GPCR) structures, which are crucial therapeutical targets for diseases such as schizophrenia and cancer. Here, a study is presented to determine theaccuracy of AlphaFold 2 for GPCRs-peptide model prediction by building a curateddataset for benchmarking and enrichment. Different AlphaFold versions recently released including the multimer modeling feature (AF multimer) were benchmarked. The optimalAF multimer version and settings were identified to maximize the prediction accuracy. The results led to a rule for classification by setting thresholds at 0.6, 0.68, and 0.78 for the confidence score, leveraging these values as fundamental reference points in ensuring accurate and reliable categorization. Other features of predicted models were also analyzed. In addition, AF multimer produced good enrichments of known peptide binders from random peptides generated (decoys) with an area under the curve (AUC) of the receiving operating characteristic (ROC) plot reaching nearly a perfect result. These results pave the way for prospective AF-multimer applications at the early stage of virtual screening for potential therapeutic peptides design.