Atlas

Benchmarks

← All benchmarks

ExploitGym v0 kernel intended-vulnerability exploits

Agents

Count of the 193 Linux-kernel tasks in ExploitGym's original v0 paper cohort exploited through the intended vulnerability with standard mitigations disabled. This cohort predates the 186-task v1 release and its extended syzbot vulnerability descriptions.

Top models (higher is better)

ModelScore
Gemini 3.1 Pro Preview0.0
Loading Atlas data…