Apple Expands Mac AI With Meta and New MLX, Running 816 Billion-Parameter Models Locally
Apple said its updated MLX framework now lets researchers run large language models locally on Apple silicon Mac Studio, extending this week’s Meta collaboration around Muse Spark Open.
Apple has expanded its Mac AI capabilities by integrating the updated MLX framework, allowing researchers to run large language models locally on Apple silicon Mac Studio computers, in collaboration with Meta.
What happened
Apple has enhanced its Mac AI capabilities by updating the MLX framework to support running large language models locally on Apple silicon Mac Studio computers. This update includes support for M5-family GPUs, RDMA over Thunderbolt, and efficient handling of complex AI architectures. Apple and Meta have collaborated on the Muse Spark Open model, which can be run on Macs with M5 Ultra processors, offering high performance and memory bandwidth.
Why it matters
This development is significant for AI developers and researchers as it enables local AI processing on Macs, reducing reliance on cloud services and offering better data control. It also highlights Apple's strategic move to position Macs as a viable alternative for advanced AI workloads.
- Apple's MLX framework now supports running large language models locally on Macs.
- Collaboration with Meta enables high-performance AI workloads on Mac Studio.
- RDMA over Thunderbolt and M5-family GPU support enhance local AI capabilities.
- Local AI processing on Macs offers better data control and lower costs.
- Apple is positioning Macs as a viable alternative to cloud-based AI solutions.
The bigger picture
The shift to local AI processing on Macs can reduce operating costs and improve data security, making it an attractive option for organizations and researchers looking to handle sensitive data without cloud dependencies.
Apple's expansion of Mac AI capabilities through the updated MLX framework and collaboration with Meta marks a significant step in shifting advanced AI workloads from the cloud to local devices, emphasizing data control and cost efficiency.
Technical details
- Support for M5-family GPU neural accelerators in MLX.
- RDMA over Thunderbolt for low-latency communication across multiple Macs.
- Efficient handling of mixture of experts (MoE) architectures.
- Muse Spark Open model running on Macs with M5 Ultra processors.
- Up to 512 GB of unified memory and 1.2 TB/s bandwidth for Meta's two M5 Ultra Mac Studios.
FAQ
What is the MLX framework?
The MLX framework is Apple's updated tool that allows researchers to run large language models locally on Apple silicon Mac Studio computers.
What is the Muse Spark Open model?
The Muse Spark Open model is a large language model developed by Meta, which can now be run locally on Apple's M5 Ultra Mac Studios.
What are the new features in the updated MLX framework?
The updated MLX framework includes support for M5-family GPUs, RDMA over Thunderbolt for low-latency communication, and efficient handling of complex AI architectures like mixture of experts.
How does Apple's strategy benefit users?
Apple's strategy to shift AI workloads to local Macs reduces reliance on cloud services, lowers operating costs, and enhances data control for users.
What performance metrics were reported for the Muse Spark Open model?
Apple's tests showed that four Mac Studios generated over 36 tokens per second for Muse Spark Open, which is faster than the average human reading pace.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by Zetik Blog on Oct 16, 2026.
Comments
Sign in to join the discussion.