Musubi Opens Up A 1.7B AI Model That Judges Content In 50 Milliseconds

Musubi said on October 6, 2026 that it has released PolicyLM-1.7B, a small decision model built for real-time content moderation and distributed with open weights, TechCrunch reported.

The model is designed to take a content policy written in ordinary English and apply it to a message in under 50 milliseconds, at roughly the cost and speed of the AI classifiers that handle moderation on most social platforms, according to the report. The pitch is flexibility: because it is built on a modern LLM architecture, it can enforce complicated rules without custom training, and policy teams can rewrite the rules without retraining the model.

"Product teams just want a better understanding of what's happening on their platform, especially as the amount of content is exponentially increasing," said Filip Jankovic, Musubi co-founder and chief AI officer, as quoted by TechCrunch.

Decision models output outcome probabilities rather than text — in this case a binary in-category or not — which makes them cheaper and faster to run than full language models. Interest in the category picked up after TypeSafe AI released its Jev model in September 2026, followed by competing products from OpenAI and Amazon. Jankovic said his own work on the approach dates to a 2024 project called GLiNER.

Musubi's announcement leans directly on that comparison, positioning PolicyLM-1.7B as a moderation-specific model customers can run themselves.

This story was first reported by TechCrunch.