Musubi, an AI company, has released a new model built to help online platforms moderate content. The model is called PolicyLM-1.7B.
The company announced the model on Tuesday, October 6. It was released with open weights, which means people can download it and run it themselves.
What PolicyLM-1.7B Does
The model takes a content policy written in plain English and applies it to messages. Musubi says it can do this in under 50 milliseconds.
Musubi designed the model to match the cost and speed of the AI classifier systems used today. Those systems power moderation on most social platforms.
The difference is flexibility. Because PolicyLM-1.7B is built like a modern large language model, it can apply complex policies without special training.
The model also does not need new training when a policy changes. This lets the people who write the rules update them as often as they need.
Filip Jankovic is Musubi's co-founder and chief AI officer. He says the tool 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 said. "Being able to label all of that in a very scalable, customizable way is extremely useful."
How Decision Models Work
Decision models do not write text like chatbots do. Instead, they output the chances of certain outcomes.
In Musubi's case, the model gives a simple yes or no answer. Either the content fits a category, or it does not.
Limiting the answer to a set of fixed choices helps these models run faster and cheaper than large language models. They still use the same transformer design that powers modern AI tools.
One early use for decision models is controlling the behavior of AI agents. Musubi is applying the same idea to posts and messages written by people.
Decision models have drawn wide attention since TypeSafe AI released its Jev model in September. OpenAI and Amazon followed soon after with competing decision models of their own.
Jankovic says his interest in this type of model came before Jev. He traces it back to a 2024 project called GLiNER, a model for finding names and key terms in text that used many of the same techniques.
Musubi is not avoiding comparisons to Jev. The company is using the new interest in decision models to draw attention to content moderation.
"If Jev caught your eye, PolicyLM-1.7B is the same kind of model, trained specifically for content moderation, that you can run yourself," Musubi's product announcement reads.
