Musubi releases open-weight PolicyLM-1.7B for real-time content moderation

On October 6, Musubi announced the release of PolicyLM-1.7B, a lightweight decision model for real-time content moderation, with open weights. The model can apply content policies written in plain English to messages in under 50 milliseconds.

Musubi says PolicyLM-1.7B is similar in cost and speed to the AI classifiers that power moderation on most social platforms. Because it retains the flexibility of a modern LLM, it can apply complex policies without special training. When policies change, the model does not need to be retrained, allowing human policy-setters to iterate as needed.

Co-founder and chief AI officer Filip Jankovic said the model gives platform managers a way to label content proactively. As content volumes grow exponentially, product teams want a better understanding of their platforms, and labeling everything in a scalable, customizable way is useful, he said.

Decision models have drawn attention since TypeSafe AI released its Jev model in September 2026, followed by competing models from OpenAI and Amazon. Unlike LLMs that output text, decision models output outcome probabilities — in this case, a binary judgment about whether content belongs to a category. By constraining outputs to predetermined choices, decision models run faster and cheaper while keeping the flexibility of transformer architecture.

Jankovic said his interest in decision models predates Jev, tracing back to a 2024 project called GLiNER that used many of the same techniques. Musubi welcomes comparisons to Jev, saying PolicyLM-1.7B is the same kind of model trained specifically for content moderation that users can run themselves.