Safety Classification on Aya Redteaming
94RecallIBM Granite Guardian 3.1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IBM Granite Guardian 3.1Size=2B, Reasoning=false2025.12 | 94 | — | |
| Qwen3GuardSize=8B, Mode=strict, Reasoning=false2025.12 | 94 | — | |
| IBM Granite Guardian 3.2Size=3B, Reasoning=false2025.12 | 93 | — | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=false2025.12 | 93 | — | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=true2025.12 | 93 | — | |
| IBM Granite Guardian 3.2Size=5B, Reasoning=false2025.12 | 89 | — | |
| AprielGuardSize=8B, Reasoning=false2025.12 | 88 | — | |
| AprielGuardSize=8B, Reasoning=true2025.12 | 88 | — | |
| ShieldGemmaSize=9B, Reasoning=false2025.12 | 78 | — | |
| Qwen3GuardSize=8B, Mode=loose, Reasoning=false2025.12 | 76 | — | |
| gpt-oss-safeguardSize=20B, Reasoning=true2025.12 | 70 | — | |
| Llama Guard 2Reasoning=false2025.12 | 64 | — | |
| Llama Guard 3Reasoning=false2025.12 | 62 | — | |
| Llama Guard 4Reasoning=false2025.12 | 44 | — | |
| CREST-BASEModel Variant=Base, Backbone=XLM-RoBERTa-Base2025.12 | — | 92.39 | |
| CREST-LARGEModel Variant=Large, Backbone=XLM-RoBERTa-Large2025.12 | — | 90.53 |