SYLIN

Muse Glimmer

by Meta · Model · Language & code models

Meta open 30B dense model, Apache 2, about 60 GB with GGUFs; AA leaderboard shows it does not beat Qwen 3.6

AccessOpen weightsNeeds~77 GB; smallest build ~21 GBDownload60 GBSize30B parametersLicenseApache-2.0

Evidence

  1. Aug 20
    Local user praises Muse Glimmer 30B for technical writing: no fluff, can ingest a repo and find relevant parts
  2. Aug 11
    30B open-weights model from Meta Superintelligence Labs under Apache 2.0, distilled from Muse Spark with a perception encoder for images
  3. Aug 11
    Quantized model generated a working HTML physics infographic with sound reasoning; code generation judged pretty good, minor animation quirks
  4. Aug 10
    Beats Qwen 27B and a second rival on agentic work (MCP orchestration, deep search, banking, long-context recall)

Cost to run

Alternatives

QwenQwen 27B still beats Muse Glimmer on computer use, terminal work and SWE-bench in Meta's table
Maxwell FlashClaimed at or near the top on reasoning and leading on agentic search, even passing proprietary models
Made from