We compared 8 mainstream synthetic accessibility scoring models against real synthesis step counts from 1.5M+ drug-like molecules - here is what we found, including the limitations of our own model.