Meta and Apple team up on AI model tailored for Mac computers - Financial Times

Meta and Apple have announced a collaboration to optimize the performance of the newly released Muse Spark Open artificial intelligence model on Apple hardware. The model, which Meta describes as the most capable open-weight AI to date, features 816 billion parameters and utilizes a mixture-of-experts architecture. The partnership focuses on leveraging Apple’s MLX framework to allow the model to run locally on desktop environments, specifically utilizing high-end Mac Studio computers.

The technical implementation allows developers to run the model on configurations ranging from two to four Mac Studio units. By enabling local execution rather than relying on cloud-based processing, the companies state that users can perform complex tasks such as software engineering, video analysis, and multi-step reasoning with increased security and reduced operational costs. Apple has released technical documentation detailing the integration of the model with its MLX array framework, which is designed to optimize machine learning tasks on Apple Silicon.

This collaboration marks a shift in the relationship between the two technology firms, which have historically maintained a competitive stance. Tensions between the companies were notably high following Apple’s 2021 privacy updates, which restricted digital advertising targeting capabilities and negatively impacted Meta’s revenue models. The current project is the result of several months of joint development efforts.

Public statements from leadership at both companies highlighted the significance of the technical achievement. Meta’s chief AI officer noted the collaborative nature of the work, while leadership from both organizations expressed support for the model’s performance on desktop hardware. The Muse Spark Open model is being released under a license that permits limited commercial use, allowing developers to download the weights and deploy the technology on their own systems.