AI & Tech · National
Meta releases Muse Spark Open, an open-weight AI model that it built with Apple to run on Mac Studio desktops
Meta launched Muse Spark Open, an open-weight AI model with 816 billion parameters that can run on two Apple M5 Ultra Mac Studios. The two companies, usually rivals, worked together on the release and are publicly backing each other.
Listen to this brief · 5 min
Read aloud by an AI voice from the text below. Download MP3
What happened
Meta on Thursday released Muse Spark Open, a new artificial intelligence model that the company calls the "most capable US open-weight model to date," The Verge reported. Meta also says it is the first model in its class that can run on a desktop setup made of two Apple M5 Ultra Mac Studios using the latest version of Apple's MLX framework.
TechCrunch reported that the model was designed to work with Macs. It is available now through a number of API platforms, according to TechCrunch. Readers following our AI and tech coverage will recognize the open-weight approach, in which a company releases a model's weights publicly rather than keeping them private.
An unusual partnership between rivals
The Verge noted that Meta and Apple, which are typically rivals, worked together on the launch. Axios reported that the companies had been collaborating for several months, according to The Verge.
What the CEOs said
Meta CEO Mark Zuckerberg wrote on X that the model "runs great on new Mac Studios thanks to work with @johnternus and Apple." New Apple CEO John Ternus replied, "Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk," The Verge reported.
Meta chief AI officer Alexandr Wang struck a lighter tone with Axios. "We're excited for mommy and daddy to be playing nice again," Wang said, as quoted by The Verge.
Apple's technical contribution
Apple also published a technical post about MLX, which The Verge described as an "array framework" for machine learning tasks on Apple Silicon. That post covered Muse Spark Open and two other open models.
What the model can do
According to TechCrunch, Muse Spark Open uses 816 billion parameters and a 1 million token context window. Meta describes it as natively multimodal and says it excels at long-horizon tasks, complex software engineering and multi-step reasoning, The Verge reported.
Meta's own test cases
Meta says it paired the model with Muse Code, its AI coding terminal, and used it in several agentic workflows, TechCrunch reported. One example involved coding work on a Mastodon client that Meta said included bug fixes, new features and a full build.
Meta also says it used the model to run DaVinci Resolve, the high-end video editing software. According to TechCrunch, Meta said the model was able to review and crop footage. TechCrunch noted that Meta's claims of better performance than other popular open-weight models are based on evaluation results the company itself released.
What it takes to run it at home
Running the model locally still requires a great deal of computing power. TechCrunch, citing Engadget, reported that the model needs at least 1 terabyte of RAM. It runs best on the Mac Studio equipped with the M5 Ultra chip, according to TechCrunch.
Why open-weight matters
TechCrunch reported that open-weight models can be significantly more cost-efficient than closed, proprietary models. They can also offer transparency and flexibility that closed source models do not allow, according to the outlet.
What to watch next
- How the model performs beyond the evaluation results Meta released itself, which form the basis of its performance claims, according to TechCrunch.
- Availability through the API platforms Meta is using to offer the model, as reported by TechCrunch.
- Apple's MLX technical post, which The Verge said also covers two other open models besides Muse Spark Open.
- Whether the hardware bar of at least 1 terabyte of RAM, reported by Engadget via TechCrunch, limits who can run the model locally.
