Muse Glimmer 30B: An On-Device Agentic Model
Published in Meta Superintelligence Labs Technical Report, 2026
Muse Glimmer is an open-source 30B-parameter model designed for autonomous agents running locally on consumer hardware. It accepts text and image inputs and combines multi-step reasoning, reliable tool use, and failure recovery. Distillation from Muse Spark, 4-bit quantization, and speculative decoding enable efficient on-device inference. The model is released under the Apache 2.0 license.
I am listed as a core contributor in Section 11.1 of the technical report (page 14).
Recommended citation: Muse Glimmer team. 2026. "Muse Glimmer 30B: An On-Device Agentic Model." Meta Superintelligence Labs. Technical report.
Download Paper