Meta Releases Muse Glimmer: 30B Open-Weight Agent Model Runs on a Laptop
Summary: Meta released Muse Glimmer, a 30B-parameter open-weight agentic model under Apache 2.0, designed to run entirely on-device with a single consumer GPU.
Key Points
- Specs: 30B dense multimodal model, 131K-token context, trained across 100+ languages; free download on Hugging Face.
- Agent-native design: Built for multi-step tasks — tool calls, code generation and debugging, local file and screenshot handling — plus LLM-as-judge use cases.
- Runs locally: Operates on a standard Mac or Windows PC with one consumer-grade GPU, no cloud required.
- Distilled from Muse Spark: Glimmer is a distilled version of Meta's flagship closed model, Muse Spark 1.2; Meta has signaled it may also open-source Spark 1.2 weights.
Why It Matters
As the AI industry moves toward on-device agent execution, Muse Glimmer gives individuals and small teams a capable, free, locally-run alternative to cloud-based agent APIs. The release also makes Meta's "personal intelligence" strategy tangible — showing a path toward AI agents that operate fully under user control, without subscription costs or data leaving the device.
More
- TechCrunch coverage — TechCrunch
- Neowin: runs locally on PCs — Neowin
- Bloomberg report — Bloomberg