> We ran a set of popular open-source models against our IDOR benchmark.
"our IDOR benchmark", there you go.