Musubi's PolicyLM-1.7B Applies Plain-English Content Policies in Under 50 Milliseconds

Musubi Unveils PolicyLM-1.7B for Real-Time Moderation

Musubi on Tuesday announced PolicyLM-1.7B, a lightweight decision model built for real-time content moderation and released with open weights. The model is designed to take a content policy written in plain English and apply it to messages in under 50 milliseconds, a speed and cost profile the company says is similar to the AI classifier systems that power moderation on most social platforms.

Because PolicyLM-1.7B has the flexibility of a modern large language model, Musubi says it can apply complex policies without special training. More importantly, the model does not need new training when the policy changes, allowing human policy-setters to iterate as much as they need.

A Binary Judgement Instead of Text

Decision models have become a hot topic in the AI world since the release of TypeSafe AI's Jev in September, which was shortly followed by competing decision models from OpenAI and Amazon. Instead of outputting text, a decision model outputs outcome probabilities. In this case, the model outputs a binary judgement: either the content is in the category or it isn't.

By limiting the model's output to a set of predetermined choices, decision models are able to run faster and cheaper than large language models while still maintaining the flexibility of the transformer architecture. One early use case has been reining in misbehavior by AI agents, and Musubi argues the same technology applies naturally to human misbehavior.

Built for Platform Teams

Musubi co-founder and chief AI officer Filip Jankovic says the product gives platform managers a way to label content proactively. "Product teams just want a better understanding of what's happening on their platform, especially as the amount of content is exponentially increasing," Jankovic says. "Being able to label all of that in a very scalable, customizable way is extremely useful."

Jankovic says his interest in decision models predates Jev, tracing it back to a 2024 project called GLiNER, short for Generalist Model for Named Entity Recognition, that deployed many of the same techniques. The product announcement frames the new model as a moderation-focused alternative to the general decision models now drawing industry attention: "If Jev caught your eye, PolicyLM-1.7B is the same kind of model, trained specifically for content moderation, that you can run yourself."

Sources: Techcrunch