PolicyLM-1.7B

PolicyLM-1.7B is a small AI model that classifies text using a supplied policy. A developer provides the rules and text to assess, and the model produces category scores that can be used in a moderation or review workflow.

Musubi Labs introduced PolicyLM-1.7B on 6 October 2026. Its model card provides downloadable weights under the Apache 2.0 licence.

The policy defines the decision

The model’s task is to apply the supplied classification rules. Changing those rules changes what the application asks it to detect. This allows a team to express its own categories and review criteria instead of relying only on a fixed, general-purpose moderation label.

Its output consists of scores for defined categories. The application still needs to decide what each score triggers, such as allowing a message, requesting review or blocking an operation. Those thresholds should be tested against labelled examples from the actual use case.

The weights can run under your own control

The 1.7-billion-parameter size and downloadable release allow teams to assess local deployment. Required memory and response time depend on the chosen runtime, precision and input length; the model card is the starting point for supported execution.

The release is useful for repeatable text decisions. Accuracy still depends on clear policies and representative evaluation, especially where similar messages need different outcomes because of context.