Muse Glimmer 30B

Last updated: August 11, 2026

Specs

Benchmarks

Strengths

• Permissive open-source licence, distilled from Muse Spark
• 30B dense: runs on a workstation, no cluster required
• Built for autonomous agents rather than chat

Weaknesses

• 131K context, well short of the 1M its parent model offers
• A distilled 30B gives up ground on the hardest reasoning

Best for

• On-premise deployments with data that cannot leave
• Agents running locally on developer machines
• Cost-sensitive workloads that still need vision

Compare with other AI models