Meta releases Muse Glimmer weights for local agents

Meta Superintelligence Labs has put Muse Glimmer, a 30-billion-parameter model, on Hugging Face under Apache 2.0, weights and docs included. The target is always-on agents running on a Mac or a PC with a single consumer GPU: the model plans tasks, calls tools, and recovers when those tools fail. It handles more than 100 languages and has switchable reasoning depth.
At full precision it needs more than 55 GB of memory. Quantization squeezes the language model below 20 GB. For the K-Quant-17GB build, Meta claims 3.1x faster decoding on an RTX 5090, 1.8x on an M5 Max, and 1.5x on an M4 Max.
Meta benchmarks the results against Gemma4-31B and Qwen3.6-27B. Mark Zuckerberg said the weights for Muse Spark 1.2 are coming soon.
Up next
Subagent coordination in Claude Code costs zero tokens
In Claude Code’s dynamic workflows, Claude writes a JS orchestration script that spawns subagents, and intermediate results stay out of the model’s context, so coordination costs zero tokens. 0xCodez’s guide: split any ‘and then’ where the next step doesn’t read the prior output.
Related stories
- 3.1% word error rate leads the streaming speech chart
- Meta’s open-weight Muse Spark Open lags Kimi K3 at coding
- Meta’s Muse Spark 1.3 lands on Google Cloud with 1M context
- Meta’s Muse Ember matches Haiku at about a third the cost
- Meta’s Muse models now hit 90% on its own memory eval
- Kimi K3.1 Pro weights land on HuggingFace at 2.8T params
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

