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

Safety Evaluation (Detailed Attack Metrics) on HarmBench

93.25PAIR

ReSA-SFT (Ours)

5.1127.992550.87573.7575Sep 15, 2025Oct 28, 2025Dec 11, 2025Jan 24, 2026Mar 8, 2026Apr 21, 2026Jun 4, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2025.09
93.2599.759799.2598.598.2593.598.7597.28----------------
2026.06
91.9--87.2-76.993.8----1.6151.622.71.311.80.937.581.31.331.588.41.320.7
2026.06
90.9--89.4-99.790----2.516.62.819.22.518.13.10.694.42.510.693.12.812.8
2025.09
89.7599.597.2598.599.2597.593.2599.2596.78----------------
2025.09
82.9999.5692.6497.6198.6594.4993.4691.3793.85----------------
2025.09
82.1399.4994.2796.5598.9992.2692.1788.893.08----------------
2025.09
79.759672.576.7598.598.2576.2594.586.56----------------
2025.09
78.7596.2583.75959787.2577.259989.28----------------
2025.09
77.5892.9162.9164.0698.4195.6280.2983.3381.89----------------
2025.09
75.59775.2581.596.7596.57696.2586.84----------------
2025.09
74.7599.258189.7597.593.7573.7510088.72----------------
2025.09
72.59973.7589.510085.2558.59884.56----------------
2025.09
71.0997.4265.5673.3998.0395.2279.3882.5782.83----------------
2025.09
70.587.7580.7592.2597.7568.56598.582.63----------------
2025.09
69.9796.7677.6192.7198.284.7377.893.0586.35----------------
2025.09
68.9998.5973.5384.596.9591.176.3491.6885.21----------------
2025.09
67.8398.7767.5782.9499.0680.9366.8590.4381.8----------------
2025.09
6599.7539.75529886.577.2584.575.34----------------
2025.09
64.6291.575.5389.7297.0671.972.3891.4681.77----------------
2025.09
64.510084.596.759779.258888.7587.34----------------
2025.09
6399.58894.2596.2570.588.2578.7584.81----------------
2026.06
60.3--87.2-37.548.8----0.656.70.922.70.967.60.676.743.10.972.359.41.357.6
2025.09
55.0486.1353.5467.8766.0971.962.925164.31----------------
2025.09
51.757761.573.754370.547.579.563.06----------------
2025.09
51.251004760.59888.572.2585.575.38----------------
2025.09
45.6980.0449.5749.8470.1955.1344.3947.955.34----------------
2026.06
41.6--46.9-28.845.3----3.472.83.468.53.469.83.482453.470.167.23.449
2025.09
39.757156.2558.7558.7552.7521.565.553.03----------------
2026.06
39.4--35.9-82.242.5----17.569.915.972.72066.917.829.385.917.82452.519.459.8
2026.06
36.9--45.9-11.330----3.876.23.469.33.181.33.192.631.32.880.560.64.455.8
2025.09
3190.7552.2577.2596.256257.57567.75----------------
2026.06
28.4--32.8-50.325----12.278.912.276.112.280.912.263.535.312.274.531.611.977
2025.09
28.2510042.571.59453597865.78----------------
2025.09
2510034.756999.53745.2573.7560.53----------------
2025.09
2475.254373.7596.7528.547.256957.19----------------
2026.06
21.3--30.9-60.629.1----6.985.36.979.36.980.56.955.37.86.992.719.17.286.5
2025.09
207124.2550.538.540.253938.2540.22----------------
2026.06
16--22.3-0.38.2----4.189.56.1854.993.44.197.803.898.120.52.987.4
2025.09
8.56814.7526.5336.7511.2527.2524.5----------------
2026.03
---------87.523.5--------------
2026.03
---------93.581.5--------------
2026.03
---------4128.5--------------
2026.03
---------33.517.5--------------
2026.03
---------6144--------------
2026.03
---------61.5--------------
2026.03
---------20--------------
2026.03
---------1.50--------------
2026.03
---------44.527--------------
2026.03
---------9574--------------
2026.03
---------32.514--------------
2026.03
---------30.514--------------
2026.03
---------45.519.5--------------
2026.03
---------31--------------
2026.03
---------0.50--------------
2026.03
---------00--------------
2026.03
---------8629.5--------------
2026.03
---------8128--------------
2026.03
---------6515--------------
2026.03
---------7316.5--------------
2026.03
---------7614.5--------------
2026.03
---------55.517--------------
2026.03
---------17.54.5--------------
2026.03
---------21.51.5--------------
2026.03
---------8526.5--------------
2026.03
---------89.573--------------
2026.03
---------144--------------
2026.03
---------176--------------
2026.03
---------65.515--------------
2026.03
---------3.51--------------
2026.03
---------20.5--------------
2026.03
---------20.5--------------