Meta Open-Weight AI Model Arrives for Macs
Meta has built a huge Meta open-weight AI model for Macs, and it is called Muse Spark Open. The company is open-sourcing the model, reinforcing its commitment to open models. This Meta open-weight AI model packs 800 billion parameters and was optimized specifically for Apple Silicon.
There is one major catch: you will need a staggering 1TB of RAM to run it locally. That requirement puts it well beyond the reach of most casual users, though it opens interesting doors for larger organizations.
What Is Muse Spark Open?
Muse Spark Open is Meta’s new open-weight model. According to the company, it is “the most capable US open-weight model to date.” Meta has been vocal about its belief that the US should not allow Chinese labs to lead on open-source AI, and this release is a direct answer to that concern.
The model was built with local performance in mind, specifically targeting Apple’s hardware. It is the first model optimized to run locally on Apple’s M5 Ultra-equipped Mac Studio machines.
Built for Apple Silicon
Apple’s unified memory architecture is central to how this works. As Meta explains, “Apple’s unified memory architecture lets the M5 Ultra’s integrated GPU use up to 512 GB per Mac Studio, so two machines can hold an entire model.”
The result, per Meta, is that “two Mac Studios behave like one supercomputer with a terabyte of unified memory.” That is the setup required to hold the full 800-billion parameter model in memory.
Mark Zuckerberg also weighed in on the release, noting that Muse Spark Open runs great on new Mac Studios thanks to work with @johnternus and Apple. The weights were released on October 15, 2026.
How It Performs
Benchmarks released by Meta suggest the model outperforms two leading Chinese models, GLM 5.3 and Kimi K3, on agentic tasks and visual reasoning. These are the kinds of tasks that matter for AI agents that need to plan, reason, and act across multiple steps.
It is worth noting that these benchmarks come from Meta itself. Independent verification will be needed before the performance claims can be fully confirmed.
The Hardware Requirements Are Steep
Running Muse Spark Open locally is not cheap. The model can run on four 256GB M5 Ultra Mac Studios, which will cost roughly $38,000. That amount of compute is squarely out of reach for the typical vibe coder.
Apple still has not announced specific pricing or a ship date for the 512GB Mac Studio. The company has only said it will be available in “late October.”
Who Is This Actually For?
For individuals, the investment is hard to justify. But for larger organizations that have been burning through their token budget, investing in a few Mac Studios might seem like a pretty good deal. That is especially true if Meta’s model performs as well as the company suggests.
The pitch is straightforward: instead of paying per token for cloud inference, an organization could buy the hardware once and run the model locally. For teams with heavy, sustained usage, that math can work in their favor.
Where to Download Muse Spark Open
Muse Spark Open is available to download now from Hugging Face. The weights were released on October 15, 2026, alongside Meta’s announcement.
If you have the hardware to run it, you can get started right away. If you do not, the release still matters as a signal about where Meta is heading with open-weight AI.
What This Means for Open-Weight AI
Meta’s release is notable for a few reasons. First, it is a direct challenge to the narrative that Chinese labs are leading on open-source AI. Second, it shows that Apple Silicon is becoming a serious target for large local models.
The 1TB of RAM requirement is a real barrier. But it also highlights how quickly local AI hardware is advancing. Two Mac Studios working together as one supercomputer is a meaningful milestone.
Whether Muse Spark Open lives up to its billing will depend on how it performs in the real world, not just in Meta’s benchmarks.
Meta’s Muse Spark Open is a bold entry into the open-weight AI race. It is the most capable US open-weight model according to Meta, and it is optimized for Apple’s M5 Ultra Mac Studio. But the 1TB of RAM requirement means it is built for organizations, not individuals.
For larger teams with the budget and the need, it could be a compelling alternative to cloud inference. For everyone else, it is a glimpse of what local AI will eventually require.
Meta continues to push its open-model strategy, and Muse Spark Open is the clearest statement of that intent yet.
