Conversational Policy Enforcement on DynaGuard Privacy Human-annotated (test)
87AccuracyGPT-5-mini
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-5-miniCategory=LLM-as-a-Judge baseline2026.04 | 87 | |
| ft-4.1-nanoCategory=Fine-tuned Model, Training Method=BARRED2026.04 | 87 | |
| ft-Qwen14BCategory=Fine-tuned Model, Training Method=BARRED2026.04 | 85 | |
| GPT-4.1Category=LLM-as-a-Judge baseline2026.04 | 82 | |
| ft-Qwen3BCategory=Fine-tuned Model, Training Method=BARRED2026.04 | 80 | |
| GPT-4.1-miniCategory=LLM-as-a-Judge baseline2026.04 | 78 | |
| OSS-Safeguard-20BCategory=Generic Guardrail Model2026.04 | 77 | |
| GPT-4.1-nanoCategory=LLM-as-a-Judge baseline2026.04 | 70 | |
| GliderCategory=Generic Guardrail Model2026.04 | 57 | |
| Qwen14BCategory=LLM-as-a-Judge baseline2026.04 | 53 |