Musubi launches open PolicyLM for real-time moderation

Musubi has released PolicyLM-1.7B, an open-weight decision model designed for real-time content moderation. The model applies plain English policies to text in under 50 milliseconds by outputting binary judgments rather than generated text, allowing it to run at the speed and cost of traditional classifiers while maintaining LLM flexibility. Co-founder Filip Jankovic states the tool enables proactive labeling without requiring retraining when policies change, positioning it as a scalable alternative to existing moderation infrastructure.