← All benchmarks

ExploitBench - GLM coverage union over three revisions

Code

Coverage on 41 tasks using the union of capabilities achieved across three revisions, then averaged across tasks. Distinct from per-trial mean progress and binary ACE.

Top models (higher is better)