Chat with Muse Glimmer 30B online

Muse Glimmer 30B is Meta Superintelligence Labs' compact multimodal model, released in August 2026 and distilled from the larger Muse Spark. It is a dense 30-billion-parameter model with open weights, and Meta tuned it specifically for autonomous agents that need to run cheaply and on modest hardware. On NinjaChat it is included in every plan.

Distillation is the point. Rather than shipping a small model trained from scratch, Meta compressed the behavior of a much larger one into a size that is practical to run everywhere. What survives the compression is the agentic behavior: following a plan, calling tools, and staying on task across a long run, which is the part smaller models usually lose first.

It reads images as well as text and calls functions, so it can look at a screenshot and then act on what it sees. The context window is about 131K tokens, smaller than the million-token models beside it, which is the honest trade for its size and speed.

Reach for it when you want an agent loop that is cheap enough to run constantly, or when you want an open-weight model you could also run yourself. When a task needs a very long context or the hardest reasoning, GLM 5.3 and Claude Opus 5 are a click away.