Musubi introduces PolicyLM-1.7B for real-time content moderation

Musubi has announced PolicyLM-1.7B, a lightweight decision model designed for real-time content moderation. The company says it released the model with open weights.

The model is intended to apply a content policy written in plain English to messages in under 50 milliseconds. Musubi says its cost and speed are designed to be similar to AI classifier systems used for moderation on most social platforms. It says the model can apply complex policies without special training and does not need new training when a policy changes.

Musubi co-founder and chief AI officer Filip Jankovic said the model could help platform managers label content proactively as the volume of material grows. He described scalable, customizable labeling as useful for product teams.

Decision models have drawn attention in AI since Typesafe AI released Jev in September, followed by competing models from OpenAI and Amazon. Unlike models that output text, decision models produce outcome probabilities. PolicyLM-1.7B instead returns a binary judgment: whether content is in a specified category. Predetermined outputs can make these models faster and less costly to run than large language models while retaining the flexibility of transformer architecture.

Jankovic said his interest in the approach predates Jev and goes back to a 2024 project called GLiNER, which used many of the same techniques. Musubi’s product announcement describes PolicyLM-1.7B as the same kind of model as Jev, trained specifically for content moderation and available for users to run themselves.