Meta Releases Muse Glimmer: Apache 2.0 Licensed 30B Local Agent Model
Meta Superintelligence Labs released Muse Glimmer on August 10, 2026—a 30-billion-parameter open-weight model distilled from its flagship Muse Spark. Released under a permissive Apache 2.0 license, Muse Glimmer is engineered to handle complex, multi-step agentic workflows directly on local developer machines without requiring active cloud connections.
Unlike standard dense chat models, Glimmer is co-trained with its agent harness specifically for tasks requiring sustained tool use, terminal interactions, code execution, and file management. The model is multimodal (supporting text and image inputs) and features native failure recovery and state preservation. On consumer hardware, the 30B model is highly optimized; when quantized to 4-bit precision, it comfortably fits within 18–20 GB of VRAM, making it runnable on standard consumer GPUs and Apple Silicon Macs via integrations with llama.cpp, MLX, and ExecuTorch.
By open-sourcing a dedicated agent model, Meta is positioning itself to compete directly in the local developer space alongside specialized agents like KAT-Coder-V2.5. The release signals a strategic pivot back toward permissive open-source models, providing developers with a privacy-preserving alternative to closed cloud orchestration suites.
