Meta Platforms Inc. (NASDAQ: META) will open-source its most powerful artificial intelligence models and introduce a new family of lightweight systems designed to run directly on consumer devices.
Chief executive Mark Zuckerberg unveiled the plans in an Instagram video and a lengthy policy essay on Monday. He said Meta will release the weights for its Muse Spark 1.2 model, allowing developers and researchers to download, study and modify the technology.
AI weights are the mathematical values that determine how a model processes information and generates responses.
The announcement marks the company’s latest effort to challenge rivals such as OpenAI and Anthropic while expanding its influence in the fast-growing AI market.
Additionally, Zuckerberg announced a new series of open-weight AI models called Muse Glimmer. The company designed those models to operate on laptops instead of relying primarily on remote cloud servers.
Meta’s shares rose about 2.1 per cent in premarket trading following the announcement. However, the stock remains down roughly 10 per cent this year as investors continue evaluating the company’s aggressive AI spending strategy.
Meta expects capital expenditures to reach as much as USD$145 billion this year. Zuckerberg has argued those investments will strengthen the company’s long-term position through its Meta Superintelligence Labs division, which it established last year.
The announcement also reflects Meta’s commitment to open-weight AI development. Unlike fully proprietary systems, open-weight models allow outside developers to access the trained parameters that make the software function, enabling broader experimentation and customization.
Meanwhile, many leading American AI companies continue emphasizing closed models. OpenAI and Anthropic generally restrict public access to their most advanced systems and their underlying weights.
Read more: Chinese AI chip companies report immense sales growth in H1
Read more: Government testing finds Anthropic AI tried to deceive humans with fake identities
Zuckerberg urges politicians to reduce regulatory barriers
Chinese technology companies have taken a different approach. Firms including Alibaba Group Holding Ltd. (NYSE: BABA) and DeepSeek have released increasingly capable open-weight models that compete with leading American AI systems in several areas.
Zuckerberg argued that American companies should lead the open-weight movement rather than allowing foreign competitors to dominate it. In his essay, he urged U.S. policymakers to reduce regulatory barriers that he believes place domestic developers at a disadvantage.
He argued that American AI developers face more restrictions on training data than many foreign competitors. Consequently, he said the United States should remove unnecessary obstacles instead of limiting access to foreign open-source models.
Zuckerberg maintained that stronger American open-weight models represent a better long-term strategy than attempting to block overseas competition. He suggested policymakers should focus on helping domestic developers build superior technology.
The Meta chief also called for changes to U.S. policy surrounding AI distillation. Distillation involves using the outputs of a sophisticated AI model to train another model with similar capabilities.
The practice remains controversial. Some American lawmakers argue that distillation can amount to intellectual property theft because it may reproduce capabilities developed through costly research.
Additionally, Zuckerberg said policymakers should reconsider rules governing how companies use data to train AI systems. He believes those changes would strengthen America’s position in the global AI race.
Industry analysts said Meta’s strategy could appeal to software developers seeking alternatives to both Chinese models and closed American systems.
Read more: South Korea’s AI bubble bursts, stock market plunges with US$2T lost
Read more: Chipmaker CXMT becomes Mainland China’s top stock in explosive IPO
Approach could help Meta compete with big-name companies
Neil Shah, co-founder of Counterpoint Research, said developers increasingly want high-quality open-weight AI that does not originate in China. He said Meta appears well positioned to satisfy that demand because most major U.S. competitors continue focusing on proprietary technology.
Muse Glimmer may also distinguish Meta from many rivals through its emphasis on on-device AI. Instead of sending requests to large cloud data centres, those models perform more computing directly on laptops or mobile hardware.
Furthermore, on-device processing can lower computing costs while improving response times. It can also reduce dependence on expensive cloud infrastructure for many everyday AI tasks.
Shah said that approach could help Meta compete more directly with Alphabet Inc. (NASDAQ: GOOGL), Microsoft Corp. (NASDAQ: MSFT) and other technology companies developing cloud-based AI services. He added that placing intelligent AI agents directly on personal devices could create a meaningful competitive advantage.
Zuckerberg also criticized what he described as growing pessimism surrounding artificial intelligence. He questioned why companies predicting severe economic disruption continue racing to build increasingly powerful systems.
However, he rejected the argument that only a small number of companies should control advanced AI because of safety concerns. He suggested concentrating that much technological power within a handful of organizations could create its own risks.
That position contrasts with comments from Anthropic chief executive Dario Amodei and OpenAI chief executive Sam Altman. Both executives have repeatedly warned that AI could reshape employment, although Altman has recently moderated some of his earlier statements.
Instead, Zuckerberg described a future where advanced AI becomes broadly available to individuals. He argued that widely distributed AI assistants could help people pursue personal goals, improve productivity and make better decisions rather than concentrating those capabilities within a few organizations.
.
