gpt-oss-safeguard-20b
🧠 AI Modelopenai
Open-weight 21B MoE safety model for content classification and LLM filtering with 131K context.
This model is designed for safety reasoning tasks. It uses a Mixture-of-Experts (MoE) architecture with 21 billion parameters, activating only a subset per token for efficiency. It is open-weight, meaning weights are publicly available for customization. Features include reasoning, temperature control, seed, stop tokens, and tool choice. It excels in content moderation, harmful content detection, and trustworthiness assessments. With a context length of 131,072 tokens, it can process long documents. Pricing is competitive for safety filtering applications.
💡Highlights
- ├─21B MoE open-weight safety model
- ├─131K token context, $0.07/M input
- └─Reasoning and tool choice support
🎯For
- ├─AI safety researchers
- ├─content moderation engineers
- └─enterprise trust and safety teams