Safety Classification on HarmBench
100RecallIBM Granite Guardian 3.2
Evaluation Results
| Method | Links | |
|---|---|---|
| IBM Granite Guardian 3.2Size=5B, Reasoning=false2025.12 | 100 | |
| Qwen3GuardSize=8B, Mode=strict, Reasoning=false2025.12 | 99 | |
| AprielGuardSize=8B, Reasoning=false2025.12 | 99 | |
| gpt-oss-safeguardSize=20B, Reasoning=true2025.12 | 99 | |
| IBM Granite Guardian 3.2Size=3B, Reasoning=false2025.12 | 98 | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=false2025.12 | 97 | |
| AprielGuardSize=8B, Reasoning=true2025.12 | 97 | |
| Llama Guard 3Reasoning=false2025.12 | 96 | |
| Qwen3GuardSize=8B, Mode=loose, Reasoning=false2025.12 | 96 | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=true2025.12 | 93 | |
| Llama Guard 4Reasoning=false2025.12 | 90 | |
| IBM Granite Guardian 3.1Size=2B, Reasoning=false2025.12 | 89 | |
| ShieldGemmaSize=9B, Reasoning=false2025.12 | 84 | |
| Llama Guard 2Reasoning=false2025.12 | 82 |