Binary Safety Classification on OpenAI Moderation (oai_safety)
76.76Macro F1Nemotron Safety Guard v3
Evaluation Results
| Method | Links | |
|---|---|---|
| Nemotron Safety Guard v3Architecture=Decoder-based, Size=7B-8B2026.05 | 76.76 | |
| PolyGuard-QwenArchitecture=Decoder-based, Size=7B-8B2026.05 | 72.77 | |
| WildGuard (vLLM)Architecture=Decoder-based, Size=7B-8B2026.05 | 71.72 | |
| PolyGuard-Qwen-SmolArchitecture=Decoder-based2026.05 | 67.91 | |
| Gliner-Guard-Omni2026.05 | 67.85 | |
| Qwen3Guard-Gen-8BArchitecture=Decoder-based, Size=8B2026.05 | 67.06 | |
| Opir-edge-multilangArchitecture=Encoder-based, Parameters=<100M2026.05 | 63.97 | |
| GLiGuard-LLMGuardrails-300MArchitecture=Encoder-based, Parameters=300M2026.05 | 63.96 | |
| Opir-multitask-multilangArchitecture=Encoder-based2026.05 | 61.26 | |
| Opir-multitask-largeArchitecture=Encoder-based2026.05 | 60.75 | |
| Opir-edgeArchitecture=Encoder-based, Parameters=<100M2026.05 | 59.86 |