Safety-Utility Trade-off Evaluation on S-Eval, ORFuzzSet, and NQ Aggregated
86.81F1 ScoreLLM-VA
Evaluation Results
| Method | Links | |
|---|---|---|
| LLM-VAModel=gemma-2, Size=9B2026.01 | 86.81 | |
| LLM-VAModel=Phi-4, Size=15B2026.01 | 85.26 | |
| LLM-VAModel=Qwen3, Size=8B2026.01 | 83.81 | |
| LLM-VAModel=Llama-3.1, Size=8B2026.01 | 81.72 | |
| VectorSteerModel=Qwen3, Size=8B2026.01 | 80.2 | |
| VectorSteerModel=Qwen3, Size=4B2026.01 | 79.79 | |
| LLM-VAModel=Qwen2.5, Size=3B2026.01 | 79.25 | |
| AlphaSteerModel=Qwen3, Size=14B2026.01 | 78.97 | |
| LLM-VAModel=Qwen3, Size=4B2026.01 | 78.22 | |
| SCANSModel=Mistral-v0.3, Size=7B2026.01 | 77.42 | |
| LLM-VAModel=Mistral-v0.3, Size=7B2026.01 | 76.56 | |
| LLM-VAModel=Qwen2.5, Size=7B2026.01 | 75.98 | |
| VectorSteerModel=Mistral-v0.3, Size=7B2026.01 | 75 | |
| AlphaSteerModel=Qwen3, Size=8B2026.01 | 74.74 | |
| VectorSteerModel=Qwen3, Size=14B2026.01 | 73.99 | |
| AlphaSteer+Model=Qwen3, Size=4B2026.01 | 73.53 | |
| AlphaSteer+Model=Qwen2.5, Size=3B2026.01 | 73.33 | |
| AlphaSteerModel=Mistral-v0.3, Size=7B2026.01 | 72.73 | |
| AlphaSteer+Model=Qwen3, Size=14B2026.01 | 72.55 | |
| SCANSModel=Qwen3, Size=4B2026.01 | 71.35 | |
| AlphaSteerModel=Qwen3, Size=4B2026.01 | 71.29 | |
| LLM-VAModel=Qwen3, Size=14B2026.01 | 71.29 | |
| AlphaSteerModel=Phi-4, Size=4B2026.01 | 71.26 | |
| AlphaSteer+Model=Mistral-v0.3, Size=7B2026.01 | 71.22 | |
| VectorSteerModel=Phi-4, Size=4B2026.01 | 71.19 | |
| AlphaSteer+Model=Qwen3, Size=8B2026.01 | 71.04 | |
| AlphaSteerModel=Qwen2.5, Size=3B2026.01 | 70.72 | |
| AlphaSteer+Model=Phi-4, Size=4B2026.01 | 69.77 | |
| LLM-VAModel=Qwen2.5, Size=14B2026.01 | 69.11 | |
| SCANSModel=Qwen2.5, Size=3B2026.01 | 68.89 | |
| LLM-VAModel=Phi-3.5, Size=4B2026.01 | 68.22 | |
| AlphaSteerModel=Qwen2.5, Size=14B2026.01 | 67.95 | |
| AlphaSteerModel=Phi-4, Size=15B2026.01 | 67.11 | |
| VectorSteerModel=Qwen2.5, Size=14B2026.01 | 67.1 | |
| VectorSteerModel=Phi-4, Size=15B2026.01 | 66.67 | |
| AlphaSteer+Model=Phi-4, Size=15B2026.01 | 66.23 | |
| AlphaSteer+Model=Qwen2.5, Size=7B2026.01 | 65.54 | |
| LLM-VAModel=Phi-4, Size=4B2026.01 | 65.45 | |
| AlphaSteerModel=Qwen2.5, Size=7B2026.01 | 64.37 | |
| AlphaSteer+Model=gemma-2, Size=9B2026.01 | 64.15 | |
| SCANSModel=Phi-4, Size=15B2026.01 | 64.1 | |
| AlphaSteer+Model=Qwen2.5, Size=14B2026.01 | 63.86 | |
| VectorSteerModel=Phi-3.5, Size=4B2026.01 | 63.8 | |
| AlphaSteerModel=Llama-3.1, Size=8B2026.01 | 63.51 | |
| SCANSModel=Qwen3, Size=8B2026.01 | 63.1 | |
| AlphaSteer+Model=Phi-3.5, Size=4B2026.01 | 62.5 | |
| AlphaSteerModel=gemma-2, Size=9B2026.01 | 62.42 | |
| SCANSModel=Qwen2.5, Size=14B2026.01 | 62.15 | |
| AlphaSteerModel=Phi-3.5, Size=4B2026.01 | 61.9 | |
| OriginalModel=Phi-4, Size=15B2026.01 | 61.82 | |
| SCANSModel=Qwen2.5, Size=7B2026.01 | 61.45 | |
| AlphaSteer+Model=Llama-3.1, Size=8B2026.01 | 61.22 | |
| VectorSteerModel=Llama-3.1, Size=8B2026.01 | 61.11 | |
| OriginalModel=Llama-3.1, Size=8B2026.01 | 61.04 | |
| OriginalModel=Qwen3, Size=4B2026.01 | 59.56 | |
| SCANSModel=Llama-3.1, Size=8B2026.01 | 59.31 | |
| OriginalModel=Phi-4, Size=4B2026.01 | 59.18 | |
| SCANSModel=gemma-2, Size=9B2026.01 | 58.9 | |
| VectorSteerModel=Qwen2.5, Size=7B2026.01 | 58.54 | |
| OriginalModel=Qwen3, Size=8B2026.01 | 57.53 | |
| OriginalModel=Qwen2.5, Size=3B2026.01 | 57.41 | |
| VectorSteerModel=gemma-2, Size=9B2026.01 | 57.14 | |
| OriginalModel=Qwen2.5, Size=14B2026.01 | 56.68 | |
| OriginalModel=Mistral-v0.3, Size=7B2026.01 | 56.35 | |
| SCANSModel=Phi-4, Size=4B2026.01 | 55.13 | |
| OriginalModel=Qwen3, Size=14B2026.01 | 53.02 | |
| OriginalModel=Qwen2.5, Size=7B2026.01 | 52.97 | |
| OriginalModel=Phi-3.5, Size=4B2026.01 | 50.73 | |
| VectorSteerModel=Qwen2.5, Size=3B2026.01 | 50.67 | |
| OriginalModel=gemma-2, Size=9B2026.01 | 49.14 | |
| SCANSModel=Qwen3, Size=14B2026.01 | 47.85 | |
| SCANSModel=Phi-3.5, Size=4B2026.01 | 47.76 |