ExploitGym (v1; 10h timeout, selected subset): leaderboard
Successful exploits using the intended vulnerability on a selected subset of ExploitGym v1, with the DoGNAVY agent and a 10-hour timeout. Scores are counts, not percentages or coverage of the full 869-task v1 suite. Other versions, time budgets and selected subsets are separate boards.
Metric: Successful intended exploits (#); v1; 10h timeout, selected subset. Source: www.cybergym.io. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 | 157 | #41 |
Interactive version: theaggregate.ai/benchmark?slug=exploitgym-v1-10h-timeout-selected-subset · How It Works · Data refreshed daily, snapshot 2026-10-28.