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

Case-level safety detection on ImpossibleBench

100ROC AUC

Monitor

4861.57588.5Apr 13, 2026
Updated 4d ago

Evaluation Results

MethodLinks
2026.04
100-
2026.04
100-
2026.04
100-
2026.04
99.4-
2026.04
99.4-
2026.04
98-0.02
2026.04
97.1-
2026.04
97.1-
2026.04
96-
2026.04
96-
2026.04
95.1-0.02
2026.04
94-
2026.04
92.9-
2026.04
920.02
2026.04
90.1-
2026.04
90-
2026.04
90-
2026.04
90-0.1
2026.04
88.3-
2026.04
88.3-
2026.04
85.80.015
2026.04
85.1-
2026.04
84.3-
2026.04
84.3-
2026.04
83-0.07
2026.04
82.7-
2026.04
82.6-
2026.04
81.1-0.16
2026.04
80.4-
2026.04
80-
2026.04
79.8-
2026.04
79.4-
2026.04
78.2-0.061
2026.04
78-
2026.04
77.2-
2026.04
74-
2026.04
64.9-
2026.04
59.8-
2026.04
50-
2026.04
50-