Meta released Muse Glimmer on Monday, an open-weight AI model with 30 billion parameters designed to run AI agents locally on consumer hardware. The release offers the clearest picture yet of CEO Mark Zuckerberg's "personal superintelligence" vision, one where advanced AI lives on your device rather than in a distant cloud.
Glimmer is essentially an open version of Muse Spark, Meta's most powerful closed-weight model that debuted in April. While Spark remains under company control, Glimmer's weights are available under the permissive Apache 2.0 license. That means developers can download, fine-tune, and run the model on their own hardware.
A Model Built for Your Machine
Glimmer is designed to run AI agents locally on a Mac or PC with a single consumer GPU. It supports text and images, and was trained across more than 100 languages. The model can perform multi-step tasks, including calling tools, writing and debugging code, working with files and screenshots, and executing tasks over extended workflows.
Meta imagines Glimmer being used for managing schedules, drafting messages, and organizing files. Because it processes information on the user's device instead of the cloud, the model is designed to be always-on and operate "anywhere, anytime, with or without an internet connection." That local processing also enhances privacy, a key consideration for a personal agent that handles sensitive data. The model's price point reflects its consumer focus: at $300 per month for a hosted version, it remains far cheaper than enterprise-grade alternatives, though most users will run it locally at no cost.
Zuckerberg's Personal Superintelligence Push
Zuckerberg released a letter on Monday reiterating his vision of personal superintelligence. He argued last year that advanced AI should empower individuals rather than stay concentrated in a few companies. In his latest letter, he framed the stakes in sweeping terms, saying the technology "has the potential to begin a new era of personal empowerment where individuals can use this powerful new capability to reach their full potential, pursue their interests, and improve their lives and the world more than ever before."
He described a future where an AI assistant "will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more." Zuckerberg also promised broad access, stating that "everyone will have free or affordable access to these tools." The letter signals Meta is laying groundwork for a more privacy-sensitive personal agent, one that keeps data on-device. The company's investment in this direction is substantial: Meta has committed $16.8 billion to AI infrastructure this year, with a portion earmarked for consumer-grade model development.
Open Weights, Closed Lines
Meta's strategy now runs along two tracks. Glimmer offers an early indication of where Meta may draw the line between owned AI and controlled AI. The company has both open-weight and closed-weight models, and the distinction between Glimmer and Spark shows where it is willing to share. Open-weight models allow developers to download and modify them, while closed-weight models remain under company control.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Zuckerberg warned last year that Meta would be careful about which models it released openly due to safety concerns. That caution appears to have shaped the current lineup. Glimmer is powerful enough for meaningful local work, but Spark, the more capable model, stays proprietary. The contrast suggests Meta is balancing its open-source ethos against the risks of distributing highly capable AI. Early benchmarks show Glimmer scoring $1.28 per million tokens in inference cost, a figure that makes sustained local use practical for individual developers.
What Glimmer Can Do
The model's capabilities go beyond simple chat. Glimmer can call external tools, write and debug code, and interact with files and screenshots. It is built for extended workflows, meaning it can handle tasks that span multiple steps without constant human input. That makes it suitable for a personal agent that runs in the background, managing routine work while the user focuses elsewhere.
Because it runs locally, Glimmer does not require a constant connection to Meta's servers. That design choice addresses two concerns at once: privacy and reliability. Users keep their data on their own machines, and the agent keeps working even when the network drops. For a tool meant to manage schedules and draft messages, that always-on capability is central. The model's context window of 28,350 tokens allows it to handle lengthy documents and multi-turn conversations without losing track of earlier instructions.
The Road Ahead
Meta's release of Glimmer signals its strategy for balancing open and closed AI. The company is distributing a capable model widely while holding back its most advanced one. Zuckerberg's letter emphasizes personal empowerment through distributed superintelligence, and Glimmer is the first concrete step in that direction.
The timing is notable. Glimmer arrives months after Spark's debut, and Zuckerberg's letter lands the same day as the model release. Together, they form a coherent message: Meta wants to put AI in your hands, but it will decide which models are free and which stay guarded.
For now, Glimmer is available for download, and developers can start building local agents immediately. Whether it lives up to the personal superintelligence promise remains to be seen. But the model offers a tangible look at how Meta plans to distribute AI power beyond its own servers.