Zuckerberg Wrote 14 Pages About Open AI. His Best AI Model Is Still Closed.

Meta released Muse Glimmer, an open-source version of its most powerful AI model, on the same day CEO Mark Zuckerberg published a 14-page essay advocating for open-source AI, yet the company's frontier model Muse Spark remains closed behind a paywall. Zuckerberg's essay criticizes Anthropic and OpenAI's controlled approach to AI, arguing that distributing intelligence is safer, but Meta's decision to keep its most competitive model proprietary highlights a gap between its stated philosophy and product strategy.

Mark Zuckerberg published a 14-page essay today about why open-source AI is the path forward for humanity. https://about.fb.com/news/2026/08/the-future-is-for-everyone/ Distribute intelligence rather than centralize it. Put the power in everyone’s hands. A new era of personal empowerment. On the same day, Meta released Muse Glimmer https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model , an open-source version of its most powerful model, Muse Spark, that anyone can download, modify, and build on for free. But the interesting part is, Muse Spark itself stays closed. You still pay to access it. The open version is nearly identical, Meta says, but the model that actually competes at the frontier, the one Zuckerberg’s essay is implicitly defending remains behind a paywall. That gap between the philosophy and the product decision is what makes today’s announcement interesting. Table of Contents What Muse Glimmer Actually Is To be clear about what Meta released today: Muse Glimmer is described as nearly identical to Muse Spark, the closed model Meta began selling access to last month. It can generate code, text, and images. Anyone can download it, modify it, and build on it without paying Meta anything. That’s genuinely useful. A freely available model at this capability level has real value for developers, researchers, and anyone building products who can’t afford frontier model API costs. The open source community will find things to do with it immediately. But Muse Spark doesn’t perform as well as leading models from Anthropic and OpenAI on coding, reasoning, and writing, according to benchmark leaderboards. Meta has positioned it as a cheaper alternative, not a better one. So what’s being distributed freely today is Meta’s most capable model, which also happens to be a step behind the actual frontier. Muse Spark stays closed. The model that costs money to access is the one Meta built when it decided open source wasn’t enough to stay competitive. That decision hasn’t changed. The essay doesn’t change it. What changed is that the open version of the second-tier model is now available, while the business model around the more competitive one remains intact. The Debate Zuckerberg Is Wading Into The 14-page essay isn’t just a product announcement dressed up in philosophy. Zuckerberg is taking a specific side in a fight that’s been building for months. Without naming them directly, the essay criticizes Anthropic and OpenAI’s position that frontier AI should be developed carefully and kept controlled. Zuckerberg’s counter is that concentrating AI power in a few companies is itself the danger that distributing it widely is both safer and more equitable. “Rather than centralizing superintelligence, we should distribute it,” he wrote. That argument would land cleaner if Meta’s most powerful model were part of what’s being distributed. Instead, Muse Spark remains centralized behind Meta’s own paywall, and Muse Glimmer, the open version is the model Meta built before it decided it needed something more competitive. The debate itself is real and worth having. Open source AI does democratize access, does accelerate development, and does give smaller players tools they couldn’t otherwise afford. Meta’s Llama models genuinely changed what was possible for developers without frontier lab budgets. That history is real and Zuckerberg’s advocacy for it isn’t entirely self-serving. But the timing of today’s release, open the second-best model, keep the best one closed, publish an essay about openness, makes the philosophy harder to separate from the strategy. You May Like: The Biggest AI Companies Are All Building Their Own Chips. That’s Not a Coincidence. https://firethering.com/ai-companies-building-their-own-chips/ The Model Nobody Is Talking About Yet Buried near the end of Zuckerberg’s essay is the detail that changes the calculus on everything announced today. Meta is working on a new model internally called Watermelon. It will be more powerful than Muse Spark. Zuckerberg did not say whether it will be open or closed. That’s the question worth watching. If Watermelon ships open source, today’s announcement looks like genuine conviction, Meta distributing its best work freely even when it costs them competitively. If Watermelon ships closed, it confirms the pattern: open source gets the model that’s no longer the competitive edge, closed gets the one that is. Meta has walked this line before. Llama was released openly when it wasn’t threatening anyone’s business model. When Meta fell behind Anthropic and OpenAI last year, the response was to build Muse Spark behind a paywall. Now that Muse Spark is established and Watermelon is in development, Muse Glimmer goes open. The philosophy stays consistent. The timing of what gets released and when tells a different story. You May Like: Best AI Coding Models for Consumer Hardware https://firethering.com/best-coding-models-consumer-hardware/ The Essay and the Product Are Two Different Things Fourteen pages is a long argument for openness from a company that kept its most competitive model closed on the same day it published them. None of this makes Muse Glimmer less useful. Developers will build on it, researchers will study it, and the open source community will do things with it Meta never anticipated. That’s how this always works and it’s genuinely valuable. But Zuckerberg’s essay frames this as a philosophical stance, distribute intelligence, don’t centralize it, the future belongs to everyone. The product decision frames it differently. The future belongs to everyone, except the part that competes directly with Anthropic and OpenAI. That part costs money. Watermelon will tell us which version of Meta we’re actually dealing with.