Apple and Meta have collaborated to optimize the Muse Spark Open model, which features 816 billion parameters, for operation on Apple hardware. The companies demonstrated that the model can run on two connected Mac Studio desktops, with a four-unit configuration providing increased processing speeds.
This partnership allows Meta to expand the adoption of its open-weight model among developers, positioning it against alternatives from OpenAI and Anthropic. For Apple, the initiative serves to promote its high-end desktop hardware by emphasizing privacy and the ability to process AI tasks locally without cloud reliance.
The move addresses growing corporate demand for cost-effective AI solutions. Companies such as Harvey, PNC Financial Services, and Siemens are exploring open-weight models to mitigate expenses associated with usage-based cloud pricing. AT&T reported a 56% reduction in costs for specific AI tasks by utilizing cheaper models.
The shift toward local AI processing is gaining momentum across the industry. Microsoft recently launched the Surface Laptop Ultra, priced from $2 599, which features an Nvidia chip designed to run large models offline. Meta also plans to integrate its Muse personal agent into the Windows ecosystem.
