Tool-agent system evaluation on Agentdojo
27.4Utility (No Attack)Qwen2.5-14B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-14BBackbone=Qwen2.5-14B2026.01 | 27.4 | 20.1 | 14.5 | |
| Meta SecAlignBackbone=Qwen2.5-14B2026.01 | 25.7 | 20.5 | 8.1 | |
| ReasAlignBackbone=Qwen2.5-14B2026.01 | 24.9 | 19.5 | 2.4 | |
| ReasAlignBackbone=Llama3.1-8B2026.01 | 10.9 | 7.1 | 0 | |
| Llama3.1-8BBackbone=Llama3.1-8B2026.01 | 6.3 | 6.3 | 0 | |
| Meta SecAlignBackbone=Llama3.1-8B2026.01 | 6.3 | 6.3 | 0 |