Artificial Intelligence
Meta Challanges OpenAI and Anthropic With Powerful New Open-Weight AI Push
Meta has announced its intention to open the weights of Muse Spark 1.2, but the model is not the same as an already-downloadable open model today. Meta is also releasing a separate model, Muse Glimmer, immediately. That distinction could matter to developers eager to experiment with Meta’s latest technology.
Meta is making its boldest move yet to shake up the artificial intelligence race. The company says it plans to open the weights of its most powerful AI model, Muse Spark 1.2, while launching a smaller model designed to run directly on consumer computers. The announcement puts CEO Mark Zuckerberg’s company squarely back in the open-weight AI fight and sends a pointed message to rivals including OpenAI and Anthropic, whose leading models remain largely closed.
Meta’s latest strategy could reshape the battle over who controls the next generation of AI.
Meta’s Most Powerful AI Model Is Going Open
Mark Zuckerberg announced the move alongside a sweeping essay outlining his vision for what he calls widely distributed AI.
Meta plans to release the weights of Muse Spark 1.2, allowing developers and researchers to access the underlying model parameters rather than relying exclusively on a closed cloud-based system. The company has positioned the move as part of a broader return to open-weight AI, following a more cautious approach with its latest flagship models.
However, there is an important distinction: Meta has announced its intention to open the weights of Muse Spark 1.2, but the model is not the same as an already-downloadable open model today. Meta is also releasing a separate model, Muse Glimmer, immediately.
That distinction could matter to developers eager to experiment with Meta’s latest technology.
Muse Glimmer Brings AI Directly to PCs
Meta’s second major announcement is Muse Glimmer, a smaller open-weight model built specifically for agentic workloads and designed to operate on a Mac or PC with a single graphics card. That is a potentially important shift.
Much of today’s most capable generative AI depends on powerful data centers. By moving more AI processing onto personal computers, smaller models can potentially reduce cloud costs, improve responsiveness, and give users greater control over their data.
Muse Glimmer is a 30-billion-parameter model released under the permissive Apache 2.0 license, according to current reports. Its focus on agentic and coding tasks gives Meta a direct opportunity to compete for developers who want powerful AI without depending entirely on proprietary cloud platforms.
View this post on Instagram
Zuckerberg’s Not-So-Subtle Swipe at OpenAI and Anthropic
The product announcement was accompanied by a much broader argument from Mark Zuckerberg: AI should not become controlled by a handful of powerful companies.
In his essay, the Meta CEO criticized the idea that advanced AI must be concentrated within a small number of organizations for safety reasons. Instead, he argued that increasingly capable AI should be distributed widely, giving individuals access to personalized AI agents.
That position places Meta on a sharply different path from companies such as OpenAI and Anthropic, which have generally kept their most advanced models closed.
Mark Zuckerberg also warned that American companies could lose ground to Chinese AI developers if U.S. policy makes open-weight development unnecessarily difficult.
Chinese companies including Alibaba and DeepSeek have aggressively pursued open-weight models, creating a growing competitive threat to Silicon Valley’s leading AI laboratories.
Meta Wants Washington to Change AI Rules
Mark Zuckerberg’s argument isn’t limited to technology.
He called for the United States to rethink policies surrounding AI training data and model distillation, a process in which knowledge or outputs from a powerful model can help train another system.
His concern is straightforward: if American developers face more restrictions than overseas competitors, open-weight AI could increasingly shift toward China.
Meta therefore wants Washington to create an environment in which U.S. open-source AI models can compete globally, rather than attempting to restrict access to foreign alternatives.
The company’s position also comes as the economics of AI become increasingly important. Businesses are looking for cheaper models, while developers want systems they can customize rather than simply rent through an API.
A Huge Bet on AI After Billions in Spending
The timing is significant for Meta.
The company has committed enormous resources to AI infrastructure and created Meta Superintelligence Labs to compete more aggressively with the world’s leading AI companies.
Meta’s shares have fallen roughly 10% this year as investors question whether the company’s enormous AI spending will translate into commercially valuable products. The latest open-weight push therefore serves two purposes: it challenges rivals and gives Meta another way to demonstrate that its AI investment is producing results.
Zuckerberg has also proposed a $1 billion fund to support communities affected by the construction of AI data centers, amid growing resistance to the infrastructure boom.
The AI Race Just Got More Open
Meta’s strategy could prove risky. Open-weight AI gives developers enormous flexibility, but it also creates questions about safety, misuse and how companies can maintain control over increasingly powerful systems.
Meta says it plans to establish a governance structure giving independent directors a role in approving safety criteria for model releases. But Zuckerberg’s larger message is unmistakable: Meta does not want the future of AI controlled exclusively by OpenAI, Anthropic, Google and a handful of other closed-model giants.
With Muse Glimmer already available and Muse Spark 1.2’s weights promised next, Meta is betting that the fastest route to AI dominance may not be building the highest wall around its technology.
It may be tearing down the wall.
