Venoms have convergently evolved in all major animal lineages and are ideal candidates to unravel the genomic processes underlying convergent trait evolution. However, few animal groups have been studied in detail, and large-scale comparative genomic analyses to address toxin gene evolution are rare. The hyper-diverse hymenopterans are the most speciose group of venomous animals, but the origin of their toxin genes has been largely overlooked. We combined proteo-transcriptomics with comparative genomics compiling an up-to-date list of core bee venom proteins to investigate the origin of 11 venom genes in 30 hymenopteran genomes including two newly sequenced genomes of stingless bees. We found a more distinct pattern in which toxin genes originated predominantly by single gene co-option, a prevalent mechanism in parasitoid wasps. These are always accompanied by parallel expansion events in other bee and hymenopteran sister gene groups. The short toxic peptides melittin and apamin appear to be unique to bees, a result supported by a novel machine learning approach. Based on the syntenic pattern we propose here Anthophilin1 as a bee-unique gene family that includes apamin and MCDP. It appears that bees co-adapted their venom genes which predominantly already existed before the aculeate ovipositor evolved to a stinger which exclusively injects venom. Our results provide insight into the large-scale evolution of bee venom compounds, and we present here the first study revealing general processes of venom evolution at a comparative genomic level for this mega-diverse Hymenoptera.