Meta (META.US) releases lightweight AI model Muse Glimmer; open-source model available for free download and can run locally on a single GPU.

According to Jinse Finance, Meta (META.US) has launched a lightweight new artificial intelligence model that is small enough to run on a single computer, allowing users to download and customize the technology.

On Monday, Meta stated that this new model, called Muse Glimmer, is a distilled (streamlined) version of the company's Muse Spark 1.2 model, with a design focus on improved efficiency to minimize system configuration requirements.

According to Meta, Muse Glimmer has 3 billion parameters and is "small enough" to run using only one GPU. Its main use cases will focus on agent-type AI tasks, such as schedule management and file organization.

The weights for Muse Glimmer (numerical parameters that help AI systems make decisions) will be available for download on Hugging Face. Meta also plans to provide weights for more powerful versions of Muse Spark. Downloads are free of charge, but users must bear their own computing costs.

This move to open source AI models via license agreements echoes China’s market competition model—in China, startups such as DeepSeek and Moonshot AI, as well as internet giant Alibaba, have already adopted this strategy to attract the largest possible user base.

In a wide-ranging essay containing policy recommendations published on Monday, Meta CEO Mark Zuckerberg stated, "Instead of centralizing superintelligence, we should distribute it widely, giving everyone the ability to control it." "This could usher in a new era of personal empowerment, enabling individuals to use this powerful new capability to fully unlock their own potential."

Contrasting Meta's approach with that of American competitors "focused on building AI for enterprises, governments, or other organizations," Zuckerberg emphasized that Meta is working to "build personal superintelligence for everyone."

The company is currently in a global race to achieve AI breakthroughs that deliver practical benefits, aiming to generate returns on its massive investments in hardware and data centers. Alongside U.S. peers such as Amazon, Google’s parent company Alphabet, and Microsoft, Meta has pledged to invest over $2 trillion to expand its computing power—a move that developing AI models and tools capable of attracting users away from competitors should help secure its long-term revenue. Last month, the company began charging developer fees for access to its Muse Spark 1.1 model for the first time.

Meta is also drawing up a cloud infrastructure business plan to offer access to AI computing power and models in a manner similar to Amazon Web Services (AWS). However, as Zuckerberg has yet to reveal more details about the plan, investors concerned about the company’s return on investment (ROI) remain rather frustrated.

On Monday, Meta announced plans to set up a $1 billion fund to invest in local U.S. communities where it owns and operates data centers. Meta, Google, and other hyperscale computing companies have faced resistance in several regions of the U.S. due to opposition over data centers' land use, energy consumption, and water usage, with their runaway capital expenditures drawing particular pushback.