Meta and Apple, typically competitors in the technology sector, have collaborated to optimize the performance of Meta’s new open-weight super intelligence (SI) model on Apple hardware. The partnership centers on enabling the Muse Spark Open model to run efficiently on desktop computers, marking a significant convergence of hardware and model development.
What Happened
Meta announced the Muse Spark Open SI model, claiming it is the “most capable US open-weight model to date.” The company stated that it is also “the first model in its class” that can “run on your desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework.” This development highlights a shift where major hardware and software vendors are working together to support large-scale SI deployments on consumer-grade desktops rather than relying exclusively on data center clusters.
Meta CEO Mark Zuckerberg publicly acknowledged the collaboration on X, stating, “Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.” John Ternus, the new Apple CEO, responded by congratulating the teams and noting, “Amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!” Additionally, Meta chief SI officer Alexandr Wang commented to Axios that the companies had been collaborating over several months, saying, “We’re excited for mommy and daddy to be playing nice again.”
According to Meta, the Muse Spark Open model is a “natively multimodal model” with 816 billion total parameters. The company reports that it “excels at long-horizon tasks, complex software engineering, and multi-step reasoning.” To support this, Apple published a technical post detailing MLX, an “array framework” for machine learning tasks on Apple Silicon, specifically covering Muse Spark Open and two other open models.
Why It Matters
This partnership signals a growing industry trend toward on-device SI inference, reducing the dependency on cloud infrastructure for high-performance models. By optimizing an 816-billion-parameter model for the Apple M5 Ultra Mac Studio, the companies are demonstrating that powerful SI tools can operate locally on desktop hardware. For developers, this means greater accessibility to large-scale SI models without the latency or cost associated with remote server calls. It also suggests that future SI model releases may be designed with specific hardware optimizations in mind, fostering tighter integration between chip makers and model developers.
The Bottom Line
Meta and Apple have jointly enabled the Muse Spark Open SI model to run on two Apple M5 Ultra Mac Studios. The collaboration leverages the MLX framework to support a natively multimodal model with 816 billion parameters, emphasizing efficient on-device performance for complex reasoning and software engineering tasks.