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

Agent Safety Reasoning on ASSEBench-Corrected

84.72Accuracy

DRAFT

30.005644.210358.41572.6197Feb 11, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.02
84.7287.8682.9293.43
2026.02
84.1787.2593.3391.55
2026.02
83.8987.3481.6393.9
2026.02
83.6187.3180.5695.31
2026.02
82.7884.7389.1280.75
2026.02
81.6786.0878.1695.77
2026.02
81.3984.0985.183.1
2026.02
81.1183.9684.3683.57
2026.02
80.5685.0478.0493.43
2026.02
80.2884.6778.4292.02
2026.02
78.6184.1974.8296.24
2026.02
78.0683.6474.8194.84
2026.02
74.1780.2573.2688.73
2026.02
73.6181.1970.2196.24
2026.02
73.6175.284.7167.61
2026.02
65.1474.666.8184.44
2026.02
64.7269.8370.6769.01
2026.02
63.3869.3470.3768.34
2026.02
61.1170.8363.6779.81
2026.02
60.867.6567.6567.65
2026.02
60.5666.5166.8266.2
2026.02
57.551.4379.4138.03
2026.02
57.153.7977.4241.21
2026.02
35.0119.8636.7113.62
2026.02
32.1118.0433.6512.33