Muse Glimmer 30B

A 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark 1.1 and optimized to run on consumer hardware.

Strengths

Handles multimodal inputs including text, images, video, audio, and documents within a 131k token context, with efficient inference suitable for running locally or on limited hardware.

Best for

Long-horizon autonomous agent tasks where a smaller, deployable model is preferred over the larger Muse Spark reasoning models.

Limitations

As a distilled model, it trades reasoning capability for efficiency compared to Muse Spark 1.1 and may not handle the most complex agentic reasoning as effectively.