Meta Brings Advanced AI Models to Apple’s Mac Computers

Meta has launched Muse Spark Open, a new open-weight artificial intelligence model designed for complex coding tasks, multi-step reasoning and AI-powered workflows. The release also marks a notable collaboration with Apple to enable the model to run locally on high-end Mac computers.

Announced on October 15, 2026, Muse Spark Open contains 816 billion total parameters, with 42 billion active parameters, and supports a context window of up to one million tokens. Meta describes it as its most capable US-developed open-weight model to date.

The launch gives developers access to model weights that they can download and run on their own infrastructure, subject to Meta’s licence terms. It also reflects growing interest in running advanced AI models locally rather than relying entirely on cloud-based services.

What Is Muse Spark Open?

Muse Spark Open is a natively multimodal AI model built to work across demanding technical and creative tasks. Its intended applications include software engineering, coding assistance, long-running agent workflows and multi-step problem-solving.

With 816 billion total parameters, the model is large, although only a portion of its parameters is active for a given operation. This design can help balance model capability with computational requirements.

Meta says it has tested the model with its Muse Code coding agent on tasks such as fixing bugs, adding features and building software. The company also demonstrated a workflow in which the model interacted with DaVinci Resolve, reviewing video footage and selecting content for vertical cropping.

How Does It Run on Macs?

Running a model of this size on a personal computer requires substantial memory and processing bandwidth.

Meta worked with Apple to optimise Muse Spark Open using MLX, Apple's machine-learning framework designed for Apple silicon. The initial local deployment is aimed at Mac Studio systems equipped with the M5 Ultra chip.

According to Meta, two appropriately configured Mac Studio computers can be connected to provide approximately one terabyte of combined unified memory, allowing the model to run across the machines.

This is a specialised setup, not a typical consumer Mac configuration. The memory requirements mean most standard laptops and desktop computers will not be able to run the full model locally.

Why Open-Weight AI Matters

Open-weight models allow developers to download and inspect a model's learned parameters, customise it for particular applications and deploy it on supported infrastructure.

This approach can offer several advantages:

  • Greater deployment flexibility: Organisations can choose compatible local hardware or hosted inference providers.

  • Customisation: Developers can adapt models to specialised tasks and workflows.

  • Local processing options: Running a model on an organisation's own infrastructure can reduce dependence on external AI services for selected workloads.

  • More control over integration: Teams can connect models with their own software and tools, subject to technical and licensing restrictions.

However, open-weight does not necessarily mean fully open-source. Meta distributes Muse Spark Open under its own Muse Model License, which includes specific conditions, particularly for certain commercial uses. Organisations should review the licence before deploying the model.

A New Chapter in the Meta-Apple Relationship

The collaboration is notable because Meta and Apple have historically competed over mobile platforms, privacy policies and control of the user experience.

Their work on Muse Spark Open shows how competition in AI is creating opportunities for cooperation. Apple’s MLX framework and unified memory architecture provide a platform for running large models locally, while Meta gains another route for making its AI technology available to developers.

The partnership could also encourage more experimentation with locally deployed AI systems for software development, research and creative production.

Availability and What Developers Should Know

Meta has released Muse Spark Open's model weights through platforms including Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Developers can also access it through several hosted inference providers.

Although the weights are available, running the complete model locally requires specialised hardware and configuration. Developers without that infrastructure can explore hosted options instead.

What This Means for the Future of AI

Meta's launch highlights two important trends in AI development: the rise of large open-weight models and the growing effort to run capable AI systems on local hardware.

Local deployment can give developers more control over their environment and workflows, but it does not automatically guarantee lower costs, stronger privacy or better performance. Those outcomes depend on hardware, configuration, licensing and the specific application.

For developers and businesses, Muse Spark Open adds another option for building AI-powered tools, particularly in coding and agentic workflows. Its demanding hardware requirements also illustrate the infrastructure challenges involved in bringing the largest models onto desktop systems.