Full-response Safety Guardrail Classification on StreamSafe internal (test)
98.7F1 ScoreSentGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| SentGuardbackbone=Qwen3-4B-Instruct-25072026.06 | 98.7 | |
| GPT-5.5protocol=zero-shot2026.06 | 95.8 | |
| WildGuard-7B2026.06 | 95.4 | |
| Qwen3Guard-8B2026.06 | 95.3 | |
| LlamaGuard3-8B2026.06 | 91.8 | |
| LlamaGuard4-12B2026.06 | 91 | |
| Gemini-3.5-Flashprotocol=zero-shot2026.06 | 87.9 |