Meta has announced an initiative to make its most advanced AI models available to the public by open-sourcing them, striving to compete with prominent entities such as OpenAI and Anthropic. During the Meta Connect event in Menlo Park, California, on September 17, 2025, CEO Mark Zuckerberg showcased the new Meta Ray-Ban Display smart glasses while sharing insights on this strategic move. Via a video on Instagram, Zuckerberg revealed plans to release the weights of the latest AI model, Muse Spark 1.2. In AI terms, "weights" refer to the underlying parameters that shape the AI's functionality and decision-making processes. Additionally, Meta is set to introduce a new series of open-source models called Muse Glimmer, which are crafted specifically for laptop use.
Meta's stock experienced a 2.1% increase in premarket trading on Monday, although shares are down approximately 10% year-to-date, raising concerns among investors about the company's spending strategies and its ability to challenge AI market leaders. Zuckerberg aims to instill confidence in shareholders regarding the progress of Meta Superintelligence Labs, established last year, despite projecting a capital expenditure as high as $145 billion this year.
The rationale behind Meta's push for open-weight AI stems from a contrasting strategy pursued by competitors like OpenAI and Anthropic, which mainly utilize closed models. In contrast, Chinese firms such as Alibaba, DeepSeek, and Moonshot have aggressively launched open-weight models that have begun to rival American technology. In a detailed 6,500-word essay discussing AI, Zuckerberg framed Meta’s open-sourcing approach as a direct challenge to Chinese advancements, urging the U.S. government to bolster domestic efforts.
Zuckerberg highlighted that foreign tech labs currently possess several advantages, principally because American companies face stricter regulations concerning training data. He insisted that U.S. policy needs to alleviate these constraints to enable American models to excel globally. "Restricting access to foreign open-source models isn't a viable solution," he stated. "We must strive for American open-source models to be the best, which is achievable only by removing the obstacles that hinder their competitiveness." Neil Shah, co-founder of Counterpoint Research, emphasized that if Western tech giants confine themselves to proprietary systems, developers will likely gravitate towards available Chinese open-weight alternatives. He noted that the demand for non-Chinese open models in the U.S. is substantial, and Meta is poised to satisfy that need.
Moreover, Muse Glimmer distinguishes itself by offering on-device AI capabilities that process data directly on consumer hardware, a departure from the reliance on expensive cloud computing infrastructures. This shift could significantly lower costs and enhance performance. Shah expressed optimism, stating that integrating smaller, efficient models like Muse Glimmer into PCs and mobile devices can help Meta effectively compete against major players like Google and Microsoft.
