List your company

Safety Alignment on HarmBench

0ASRNo Steering
-3.520.12543.7567.375Feb 24, 2025Apr 21, 2025Jun 16, 2025Aug 11, 2025Oct 6, 2025Dec 1, 2025Jan 27, 2026

Evaluation results

MethodLinks
No SteeringModel=Qwen2.5-1.5B2026.010
No SteeringModel=Qwen2.5-3B2026.010
No SteeringModel=Qwen2.5-7B2026.010
No SteeringModel=gemma-2-2b2026.010
ActAddModel=gemma-2-2b2026.010
SASModel=gemma-2-2b2026.010
No SteeringModel=gemma-2-9b2026.010
ActAddModel=gemma-2-9b2026.010
SASModel=gemma-2-9b2026.010
No SteeringModel=Llama-3.2-3B2026.011.92
LED-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.024
No SteeringModel=Llama-3.1-8B2026.015.77
No SteeringModel=Llama-3.2-1B2026.016.73
DirAblModel=gemma-2-9b2026.0111.54
ActAddModel=Qwen2.5-1.5B2026.0113.46
SASModel=Qwen2.5-1.5B2026.0113.46
LED-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0213.75
LED-MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.0214.75
LED-MergingLM=true, Math=false, Code=true2025.0214.75
w/o MergingWizardLM-13B=false, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.0215.5
LED-MergingModels (LM)=true, Models (Math)=true2025.0216
Model StockLM=true, Math=false, Code=true2025.0217.25
Model StockWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0220
LED-MergingLM=true, Math=true, Code=true2025.0220.75
LED-MergingLM=true, Math=true, Code=false2025.0221
w/o MergingLM=true, Math=false, Code=false2025.0221.5
Model StockLM=true, Math=true, Code=true2025.0223.25
DirAblModel=Qwen2.5-1.5B2026.0125
DirAblModel=gemma-2-2b2026.0125
Task ArithmeticLM=true, Math=true, Code=false2025.0226.5
w/o MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=false2025.0228.25
Task ArithmeticWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0231.75
Task ArithmeticLM=true, Math=true, Code=true2025.0232
DirAblModel=Llama-3.1-8B2026.0132.69
BreadcrumbsLM=true, Math=true, Code=false2025.0233
Task ArithmeticWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.0235
Ties-MergingLM=true, Math=true, Code=false2025.0235.75
Model StockLM=true, Math=true, Code=false2025.0236
Task ArithmeticWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.0237
Ties-MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.0237
BreadcrumbsWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.0237.75
Task ArithmeticLM=true, Math=false, Code=true2025.0238
Ties-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.0238.25
BreadcrumbsLM=true, Math=true, Code=true2025.0238.25
AASModel=Qwen2.5-1.5B2026.0139.42
Ties-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0239.5
BreadcrumbsLM=true, Math=false, Code=true2025.0239.5
Ties-MergingLM=true, Math=true, Code=true2025.0241.25
w/o MergingLM=false, Math=true, Code=false2025.0242
BreadcrumbsWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.0244.75
Ties-MergingLM=true, Math=false, Code=true2025.0245
w/o MergingWizardLM-13B=false, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0245.75
w/o MergingModels (LM)=false, Models (Math)=true2025.0246
BreadcrumbsWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.0247.25
ActAddModel=Qwen2.5-3B2026.0150.96
SASModel=Qwen2.5-3B2026.0150.96
DirAblModel=Qwen2.5-3B2026.0152.88
DirAblModel=Llama-3.2-3B2026.0153.85
w/o MergingModels (LM)=true, Models (Math)=false2025.0254.75
DirAblModel=Llama-3.2-1B2026.0154.81
Task ArithmeticModels (LM)=true, Models (Math)=true2025.0255.75
DirAblModel=Qwen2.5-7B2026.0155.77
w/o MergingLM=false, Math=false, Code=true2025.0261.25
Ties-MergingModels (LM)=true, Models (Math)=true2025.0262
BreadcrumbsModels (LM)=true, Models (Math)=true2025.0263.75
AASModel=gemma-2-9b2026.0167.31
Selective SteeringModel=gemma-2-9b2026.0168.27
ActAddModel=Llama-3.2-1B2026.0170.19
SASModel=Llama-3.2-1B2026.0170.19
AASModel=Qwen2.5-3B2026.0170.19
ActAddModel=Llama-3.1-8B2026.0174.04
SASModel=Llama-3.1-8B2026.0174.04
Selective SteeringModel=Qwen2.5-1.5B2026.0174.04
AASModel=gemma-2-2b2026.0174.04
AASModel=Llama-3.2-1B2026.0176.92
AASModel=Llama-3.1-8B2026.0177.88
Selective SteeringModel=Llama-3.1-8B2026.0177.88
Selective SteeringModel=Llama-3.2-1B2026.0179.81
ActAddModel=Llama-3.2-3B2026.0182.69
SASModel=Llama-3.2-3B2026.0182.69
Selective SteeringModel=gemma-2-2b2026.0182.69
AASModel=Llama-3.2-3B2026.0184.62
Selective SteeringModel=Qwen2.5-3B2026.0184.62
Selective SteeringModel=Llama-3.2-3B2026.0185.58
ActAddModel=Qwen2.5-7B2026.0186.54
SASModel=Qwen2.5-7B2026.0186.54
AASModel=Qwen2.5-7B2026.0187.5
Selective SteeringModel=Qwen2.5-7B2026.0187.5