Meta has released Muse Glimmer, a 30-billion-parameter open-weight AI model designed to run autonomous agents locally on consumer hardware. The model can operate on a Mac or PC with a single consumer GPU, giving developers a way to build agents that handle multi-step tasks without relying entirely on cloud infrastructure.
Meta released the model weights under the Apache 2.0 license, allowing developers to download and modify them. In its official announcement, the company said Glimmer is optimized for long-running local workflows involving tool use, coding, files, screenshots, and other tasks.
Glimmer accepts both text and images and was trained using data from more than 100 languages. Meta says the model can recover from failed tool calls, sustain reasoning across extended workflows, and work with existing agent orchestration systems.
Meta Targets Personal Agents That Run Locally
Meta describes Glimmer as suitable for personal agents that manage schedules, draft messages, organize files, and work with other private information. Running those tasks on the device can reduce the amount of personal context that needs to be sent to remote servers.
The model is designed to remain available with or without an internet connection. Meta compressed its weights to roughly 4-bit precision, bringing the language model below 20 GB so it can fit within a 24 GB or 32 GB memory envelope alongside components needed for image processing and faster generation.
Glimmer was trained partly by distilling knowledge from Muse Spark, Meta’s larger model. Muse Spark remains closed-weight, while developers can download Glimmer, fine-tune it, and run it on their own systems.
Release Fits Zuckerberg’s Personal Superintelligence Plans
The release comes alongside a new statement from Meta CEO Mark Zuckerberg describing the company’s plans to distribute advanced AI capabilities widely. Zuckerberg wrote that Meta wants people to have personal agents that can work continuously across areas including work, health, finances, relationships, and home management.
He also said Meta intends to provide free versions of its AI tools to billions of people, while offering additional compute to paying users. His statement argues that advanced AI should be distributed among individuals rather than concentrated within a small number of institutions.
Meta is still distinguishing between models it releases openly and those it keeps under tighter control. Glimmer’s open-weight release gives developers direct control over a smaller local model, while Muse Spark remains available only under Meta’s own terms.
Meta said Glimmer is available now through Hugging Face, with support planned through tools and frameworks including Ollama, LM Studio, llama.cpp, ExecuTorch, MLX, vLLM, and SGLang. The company is also working with AMD, Arm, Dell, Intel, and Nvidia on hardware optimization.
Featured image credits: TwoTone Creative
For more stories like it, click the +Follow button at the top of this page to follow us.