Table of Contents
Meta has released Muse Spark Open, a huge open-weight AI model that it calls “the most capable US open-weight model to date.” The company built it with Apple, so developers can run it on their own desks using linked Mac Studio computers. However, the setup needs about 1TB of combined memory, so most people will try it through cloud partners instead.
Key Takeaways
- Meta released the weights for its new model on Oct. 15, 2026. It has 816 billion total parameters, with 42 billion active per token.
- Meta and Apple worked together for months so the model runs locally on M5 Ultra Mac Studios.
- You need two 512GB Mac Studios or four 256GB machines, for about 1TB of unified memory in total.
- Seven API partners, including Groq, Together AI and OpenRouter, host it at launch.
- It ships under Meta’s Muse Model License, which has limited commercial terms.
What Is Muse Spark Open?
Muse Spark Open is the open-weight version of Muse Spark, Meta’s agent and coding model. In other words, anyone can download its trained weights and run the model on their own hardware.
Meta says the model has 816 billion total parameters. Yet it activates only about 42 billion of them for each token. Engineers call this design a “mixture of experts.” It also takes text, image, video and audio input, and it holds a context window of 1 million tokens.
According to Meta, the model “excels at long-horizon tasks, complex software engineering, and multi-step reasoning.” For example, Meta says the model shipped real fixes to Ice Cubes, an open-source Mastodon app. It also had the model edit video in DaVinci Resolve from start to finish.
Is it Meta’s best model?
No. Meta AI chief Alexandr Wang told CNBC, “These aren’t the most capable models that we’ve developed or released.” Instead, Meta’s top model is Muse Wonder, a proprietary system it released just days earlier. Still, Engadget notes that Meta’s own benchmarks show the open model beating two leading Chinese models, GLM 5.3 and Kimi K3, on agentic tasks and visual reasoning.
Why Meta Built Muse Spark Open With Apple
The launch is also notable because Meta and Apple have long been rivals. CNBC reports that the two firms have not struck a public partnership since 2017. Their ties hit a low after Apple’s 2021 privacy changes, which Meta said would cut its 2022 revenue by $10 billion.
Now the mood has changed. Mark Zuckerberg posted on X that the model “runs great on new Mac Studios thanks to work with @johnternus and Apple.” In reply, new Apple CEO John Ternus called it “amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk.”
Wang was more playful. “We’re excited for mommy and daddy to be playing nice again,” he told Axios, as quoted by The Verge. Meanwhile, Ternus took over from Tim Cook as Apple’s CEO last month, CNBC reports.
What Apple gets out of it
For Apple, the deal shows off its chips. CNBC says it could signal Apple’s growing strength in AI hardware, a market Nvidia has dominated. Apple also published benchmarks and a technical post about MLX, its machine learning framework for Apple silicon.
How Much Memory Does Muse Spark Open Need?
This is the catch. The model is far too large for one computer, so it must be split across several Macs.
The Mac Studio setup
Meta’s developer guide lists two supported Mac setups:
- Two Mac Studios with the M5 Ultra chip and 512GB of unified memory each
- Four Mac Studios with the M5 Ultra chip and 256GB of unified memory each
Either way, you get about 1TB of memory in total. The Macs connect directly over Thunderbolt 5, and they need macOS 26.2 or later. As a result, the machines share the work on every token.
Meta explains the trick this way: “two Mac Studios behave like one supercomputer with a terabyte of unified memory.”
What it will cost
Unfortunately, the hardware is not cheap. CNBC reports that a Mac Studio with an M5 Ultra and 256GB of memory starts at $10,799. Engadget estimates that four of them will cost roughly $38,000. Meanwhile, the 512GB model is due in late October, and Apple has not announced its price.
License and API Partners
Most developers will not buy a stack of Macs. Instead, they can call the model through a hosted service.
Where you can use it
Meta lists seven API platforms at launch: Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The weights are also available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models.
In addition, the model works with popular tools such as vLLM, SGLang, llama.cpp and Apple’s MLX. Meta says it includes optimizations for chips from AMD, Apple, Intel and Nvidia.
What the license allows
Meta released the weights under its Muse Model License. The company says the license has “limited commercial terms, which allow many developers to use the model for free.” Even so, it is a custom Meta license, so businesses should read the terms before they build on it.
What Muse Spark Open Means for the AI Race
The release matters beyond Apple fans. First, it pushes back on Chinese open models. Engadget notes that Meta has said the US should not let Chinese labs lead in open-source AI.
Second, it reflects a growing interest in running powerful AI locally, CNBC says. Local models keep data on your own machines. Both companies said that adds security and saves money. If you follow AI stocks, our report on the Nasdaq’s recent AI sell-off explains why investors are watching the sector so closely.
Finally, Meta also says it added extra safety training because open weights carry their own risks. A full evaluation report is coming soon. For another closely watched tech story, see our report on the Tesla Roadster demo.
Frequently Asked Questions
What is Muse Spark Open?
It is Meta’s open-weight AI model, released on Oct. 15, 2026. It has 816 billion parameters and targets coding and AI agent tasks.
Can I run it on a regular Mac?
No. You need two M5 Ultra Mac Studios with 512GB of memory each or four with 256GB each. That adds up to about 1TB.
Is the model free to use?
Meta says its license lets many developers use it for free. However, the terms limit some commercial uses, so check the license first.
Where can developers try it without buying hardware?
Hosted options include Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
Sources
- CNBC – Apple and Meta team up to make Muse AI model run on Mac computers
- The Verge – Meta and Apple team up to make Muse AI models run better on Macs
- Engadget – Meta built a huge open-weight AI model for Macs, but you’ll need 1TB of RAM to run it
- Meta AI Research – Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date
- Meta for Developers – Build with Muse Spark Open
- Meta Model API – Muse Spark Open on Mac
