Safety Evaluation on RRB (Red Teaming Resistance Benchmark)
2.68Hp ScoreSafesearch(79k+12k)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Safesearch(79k+12k)Backbone=Qwen2.5-7B-Instruct2026.05 | 2.68 | 18.7 | — | |
| COMPASS(8k+2k)Backbone=Qwen2.5-7B-Instruct2026.05 | 2.63 | 14.1 | — | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 2.55 | 24.7 | — | |
| Naive RAGBackbone=Qwen2.5-7B-Instruct2026.05 | 2.44 | 47.6 | — | |
| COMPASSEvaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.38 | 19.7 | — | |
| Base AgentBackbone=Qwen2.5-7B-Instruct2026.05 | 2.34 | 37.3 | — | |
| SafesearchEvaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.33 | 20.8 | — | |
| Naive RAGBackbone=Llama3.1-8B-Instruct2026.05 | 2.31 | 31.9 | — | |
| Search-R1(79k)Backbone=Qwen2.5-7B-Instruct2026.05 | 2.28 | 44.4 | — | |
| Naive RAGEvaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.26 | 41.4 | — | |
| BaseEvaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.25 | 22 | — | |
| Base AgentEvaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.25 | 44.5 | — | |
| Search-R1Evaluated Model=Qwen2.5-7B-Instruct, Judge=Qwen3-max2026.05 | 2.23 | 50.7 | — | |
| Base Agent + Document FilteringBackbone=Qwen2.5-7B-Instruct2026.05 | 2.23 | 34.1 | — | |
| Base Agent + Query FilteringBackbone=Qwen2.5-7B-Instruct2026.05 | 2.21 | 32.2 | — | |
| Naive RAGJudge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.19 | — | 40.8 | |
| Base AgentJudge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.15 | — | 48.9 | |
| BaseJudge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.14 | — | 24.8 | |
| SafesearchJudge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.13 | — | 24.5 | |
| Safesearch(79k+12k)Backbone=Llama3.1-8B-Instruct2026.05 | 2.08 | 18.6 | — | |
| COMPASSJudge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.08 | — | 22.9 | |
| Search-R1Judge=Claude Sonnet 4.5, Base Model=Qwen2.5-7B-Instruct2026.05 | 2.06 | — | 53.4 | |
| COMPASS(8k+2k)Backbone=Llama3.1-8B-Instruct2026.05 | 2.01 | 7.6 | — | |
| BaseBackbone=Llama3.1-8B-Instruct2026.05 | 1.92 | 10.2 | — | |
| Search-R1(79k)Backbone=Llama3.1-8B-Instruct2026.05 | 1.89 | 49.1 | — | |
| Base Agent + Query FilteringBackbone=Llama3.1-8B-Instruct2026.05 | 1.67 | 15.1 | — | |
| Base AgentBackbone=Llama3.1-8B-Instruct2026.05 | 1.66 | 16.4 | — | |
| Base Agent + Document FilteringBackbone=Llama3.1-8B-Instruct2026.05 | 1.63 | 15.5 | — |