Muse Glimmer 30B (TR)

Muse Glimmer 30B is a 30-billion-parameter open-weight multimodal model positioned as a lighter sibling distilled from Meta's larger Muse Spark, built with always-on local agents, tool use, coding assistance, and image understanding in mind. According to aggregator listings, the model sits within the broader muse family and is reported at approximately 29.8B parameters, which suggests a moderate-scale design aimed at fitting onto accessible local hardware rather than competing at the frontier scale of its parent. Its open-weight release makes it suitable for self-hosted deployments and gateway routing, giving developers flexibility to run inference where latency, privacy, or cost matter more than absolute peak capability.

In practice, Muse Glimmer 30B supports text and image inputs while returning text outputs, combining multimodal perception with reasoning, tool calling, structured output, attachments handling, temperature control, and interleaved thinking. Both source pages agree on a 131,072-token context window and a matching maximum output ceiling, giving the model room for long-document reasoning and extended multi-step agent workflows. The design intent of being a distilled, locally deployable multimodal variant makes it a reasonable fit for developers building lightweight agent stacks, coding copilots, and image-aware assistants who want open weights and a generous context without the compute footprint of a full-scale frontier model.