Meta has released Muse Glimmer, a 30-billion parameter AI model designed to run directly on consumer laptops and PCs, marking the first model in its Muse family to ship with open weights under the permissive Apache 2.0 licence. The model comes from Meta Superintelligence Labs (MSL) and is aimed at developers building AI agents that need to work without cloud connectivity.
Muse Glimmer is compact enough to run entirely on a single consumer GPU or a supported Mac device, a design choice Meta says allows developers to build AI experiences that function anywhere, including offline environments — a consideration increasingly relevant for developers in Southeast Asia building for markets with uneven connectivity.
Built for agentic workflows and local inference
The model is optimised for agentic workflows, coding, tool use, multimodal understanding and long-context reasoning, and is designed to complete complex, multi-step tasks by planning, using tools, checking its own results and recovering from errors. Key technical details include:
- 30-billion parameter model capable of running on a single consumer GPU or supported Mac
- Open weights released under the Apache 2.0 licence, alongside developer documentation and inference code
- Compatible with popular frameworks including Hugging Face, Ollama, LM Studio, llama.cpp, MLX, vLLM, Together AI and Fireworks AI
- Supports text and image inputs, advanced reasoning, tool use and more than 100 languages
Part of Meta’s continued open-source AI push
The release includes both full-precision and quantised model weights, as well as speculative decoding support, giving developers flexibility to balance performance against hardware constraints.
Muse Glimmer represents Meta’s latest step in advancing open AI development, giving developers greater access to powerful local AI capabilities while continuing the company’s long-standing commitment to open-source innovation, according to Meta’s announcement.
The local-first design also has implications for enterprises with strict data residency or privacy requirements, since inference can run entirely on-device without sending data to external servers — a factor increasingly weighed by regulated industries evaluating agentic AI tools in the region.
