Musubi has unveiled PolicyLM-1.7B, a lightweight decision model built specifically for real-time content moderation. Announced Tuesday, the model comes with open weights and is designed to evaluate messages against a content policy written in plain English, returning a binary judgment within 50 milliseconds. The company says it runs at a cost and speed comparable to the AI classifiers most social platforms already use, but with the flexibility of a modern LLM—so it can handle complex policies without special training.
The key selling point is adaptability. Because the model applies policies directly as written, it does not need to be retrained when those policies change. Musubi co-founder and chief AI officer Filip Jankovic says this lets product teams label content proactively, especially as content volumes grow exponentially. "Product teams just want a better understanding of what's happening on their platform," he said, describing a scalable, customizable way to label everything.
Decision models have drawn attention since TypeSafe AI released Jev in September, followed by competing models from OpenAI and Amazon. Instead of generating text, they output probabilities across predetermined choices, making them faster and cheaper than general LLMs. Musubi says its interest actually predates Jev, tracing back to a 2024 project called GLiNER, and it welcomes the comparison. The company's announcement frames PolicyLM-1.7B as the same kind of model, trained specifically for moderation, that users can run themselves.