Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Vulnerability Reproduction on CyberGym (evaluation)

1,600Total Rollouts

GPT-5.4 mini Best-of-8

518.4799.21,0801,360.8Jul 2, 2026
Updated 23d ago

Evaluation Results

MethodLinks
2026.07
1,6007932138542.535.9
2026.07
1,600910005864336.3
2026.07
1,6004491201266356.1
2026.07
1,60063511714170.563.8
2026.07
1,6006771710954.547.6
2026.07
1,6002622121226154.1
2026.07
1,026086081065346.1
2026.07
924099011904538.3
2026.07
89122640121206053.1
2026.07
842056281346760.2
2026.07
7530280181547770.7
2026.07
75005113111758.551.6
2026.07
71902119181427164.4
2026.07
61913402014572.565.9
2026.07
56011015516984.578.8