Meta launches new AI model as Zuckerberg lays out vision for personal 'superintelligence'

Meta will focus on making personal superintelligence available to billions of people, with free or affordable access, as it seeks to put advanced AI capabilities directly in the hands of individuals, founder and CEO Mark Zuckerberg said on Monday.

In a blog titled “The Future is for Everyone”, Zuckerberg said Meta’s approach to superintelligence would be based on individual empowerment, invention and a balance of power, arguing that advanced AI should not be concentrated among a small number of companies, governments or institutions.

“The defining questions of our age are who will have access to superintelligence and what will we direct it towards,” Zuckerberg said.

He said Meta expects people in the coming years to use superintelligence to create businesses, develop ideas, learn new concepts and advance health and quality of life.

His statement comes as Meta Superintelligence Labs announced Muse Glimmer, a 30-billion-parameter open-weight AI model designed to run locally on consumer devices and support “always-on” agentic workflows.

Meta said Muse Glimmer can run on a Mac or PC with a single consumer GPU and is aimed at applications including local AI agents, function calling, coding and model evaluation.

The company has released the model weights under the permissive Apache 2.0 licence and said developers can download them from Hugging Face.

Open-weight models can generally be more affordable to use than leading proprietary models developed by frontier AI companies such as OpenAI and Anthropic. They give developers access to the model's core components for greater customisation.

Zuckerberg said Meta is working towards giving everyone an “exceptionally capable personal agent” that can understand an individual’s goals and interests and work on their behalf across areas such as relationships, health, career, finances, home management and hobbies.

The personal agents, he said, would have strong privacy and security options and could be accessed through different devices, including smart glasses.

Meta also plans to provide AI-powered tools for creation, business building and learning, he said.

Zuckerberg said superintelligence could give individuals the ability to turn ideas into products and businesses without needing to raise money or build large teams.

He also envisaged AI-powered personalised tutors that could help people learn new skills, languages and subjects, while AI-driven scientific research could accelerate drug discovery and the development of personalised therapies.

Meta will offer free versions of its superintelligence tools to billions of people, while users seeking additional computing resources will be able to pay through a dynamic auction mechanism, he said.

Zuckerberg argued that invention, rather than automation, should be the primary contribution of superintelligence.

While acknowledging concerns over job displacement, he said AI could ultimately lead to job growth if individual capabilities and demand for new skills expand alongside automation.

“Which outcome we get depends on the balance in progress between automation on one side and individual empowerment and invention on the other,” he said.

Zuckerberg also addressed risks related to cybersecurity, bioterrorism, government surveillance and the possibility of AI systems becoming too powerful.

He rejected the idea that a single superintelligent system could be aligned with the competing values of all of humanity, asserting that safety should come from a balance of power in which individuals, businesses, governments and multiple AI systems can check one another.

“The solution is to ensure that superintelligence is broadly distributed to empower people,” he said.

Zuckerberg argued that widespread access to advanced AI could strengthen defences by enabling more people and organisations to identify vulnerabilities and secure systems.

He proposed that frontier AI companies provide governments with intermediate training checkpoints of advanced models and technical staff so that critical infrastructure can be hardened against emerging threats.

On biological and chemical risks, he called for continued work by AI labs to reduce risks while urging governments to focus on controlling the physical production and distribution of harmful materials and speeding up the development and approval of new treatments.

Zuckerberg also called for greater cooperation between frontier AI labs and governments, and warned against policies that could slow US AI development.

He said maintaining US leadership in AI would require faster development of energy capacity and data centres, as well as continued investment in open-source AI.

Meta, he said, would continue supporting open-source AI and plans to resume releasing some open-source models soon.

The company is also implementing an independent governance structure under which its board will approve safety criteria for model releases and review whether individual releases meet those criteria, Zuckerberg said.

He said Meta's long-term goal was to ensure that superintelligence remains aligned with individual users' goals rather than imposing a centralised set of values.

“Superintelligence holds the promise of giving everyone that power, and building a positive future for everyone,” Zuckerberg said.