Meta has released Muse Glimmer, a 30-billion-parameter multimodal model distilled from Muse Spark and tuned for always-on local agent workflows. The model is available under the Apache 2.0 license and runs on a single consumer GPU or Mac without network calls.

  • Weights are compressed to approximately 4-bit precision, allowing the model to fit into 24 GB VRAM with only 1.0% accuracy degradation.
  • DFlash block-diffusion drafting predicts 16 tokens per pass, delivering a 3.1x speedup on an RTX 5090 compared to standard decoding.
  • Muse Glimmer leads Gemma4-31B and Qwen3.6-27B on benchmarks including MCP Atlas (75.5), DeepSearch QA (74.6), and SWE-Bench Pro (51.2).
  • The model trails Qwen3.6-27B on OSWorld-Verified and TerminalBench 2.1 but maintains strong performance in agentic orchestration and reasoning tasks.

This release enables self-hosted, offline operation for regulated enterprises and developers who require data residency and low latency without per-token billing.