Meta open-sources Muse Glimmer, a 30B local AI model for always-on agents on consumer GPUs

Meta Superintelligence Labs introduced Muse Glimmer, a 30B-parameter model released with open weights under Apache 2.0. The model is optimized for local agent workflows such as tool calling, coding, function execution, and LLM-as-a-judge evaluation, with support for Mac or PC setups using a single consumer GPU.

The release reflects a broader shift toward capable local AI agents that can run offline, preserve more user context on-device, and reduce reliance on cloud infrastructure. By pairing open weights with developer integrations like llama.cpp, MLX, and ExecuTorch, Meta is trying to accelerate an ecosystem where smaller specialized models become practical building blocks for personal and enterprise automation.