← All benchmarks

ExploitGym v0 (898-task paper cohort)

Safety

The original ExploitGym paper cohort of 898 tasks: 520 userspace, 185 V8 and 193 Linux-kernel vulnerabilities. Pass@1 is the percentage exploited through the intended vulnerability with standard mitigations disabled. This v0 snapshot predates the v1 removal of 29 cases and changes to vulnerability descriptions and V8 flags, so it has a separate factor group from the 869-task release.

Top models (higher is better)