← All benchmarks

VulnPR-100

Code · 2026-10-24

BugBunny's VulnPR-100 gives an agent 100 pull requests that each introduce a known vulnerability and one hour per pull request to find it; the score is the share of the 100 vulnerabilities found, in percent. Results are published as posts by BugBunny's founder without a public task list, harness description or per-model settings.

Top models (higher is better)