As decision models spread across the industry, Musubi is applying them to content moderation. The company announced PolicyLM-1.7B, a lightweight decision model with open weights that takes a content policy written in plain English and applies it to messages in under 50 milliseconds. Musubi says the model is designed to be similar in cost and speed to the AI classifier systems that currently power moderation on most social platforms, while keeping the flexibility of a modern LLM so it can apply complex policies without special training.

A key property is that the model does not need to be retrained when a policy changes. That lets human policy-setters iterate as much as they need, and co-founder and chief AI officer Filip Jankovic frames the result as proactive labeling: product teams want to understand what is happening on their platform as content volume grows exponentially, and a scalable, customizable labeling system is extremely useful.

The release lands at a moment when decision models are getting wider attention. Typesafe AI released Jev in September, followed by competing decision models from OpenAI and Amazon. Unlike an LLM, a decision model outputs outcome probabilities rather than text; in this case, the output is a binary judgment about whether content falls into a given category. Limiting output to predetermined choices makes these models faster and cheaper than general LLMs while retaining the flexibility of the transformer architecture.

One early use case has been moderating the behavior of AI agents, and Musubi sees the same approach as a natural fit for human content. Jankovic notes his interest predates Jev, going back to the 2024 GLiNER project for named entity recognition, which used many of the same techniques. Musubi is not avoiding the comparison. The product announcement says, “If Jev caught your eye, PolicyLM-1.7B is the same kind of model, trained specifically for content moderation, that you can run yourself.”