Meta publishes Muse Glimmer 30B, an open-weight agent model for consumer hardware

Meta has published a technical report on Muse Glimmer, a 30B open-weight model built for agentic tasks such as tool use and multi-step planning.

The report covers the architecture, distillation recipe, and quantization. Meta says the model runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision, and leads comparably sized open models on tool use and multi-step planning.

A post by Dongxi NLP, quoting Meta's announcement, adds figures that Meta's quoted text does not include: 4-bit quantization brings the model to about 17GB for 24GB GPUs, and DFlash speculative decoding reaches about 233 tokens/s on an RTX 5090, which the post says is roughly 3.1 times standard decoding. The post also says some metrics exceed Qwen3.6-27B and Gemma4-31B. These are the post's claims, not independently verified results.