Meta and Apple are working together on artificial intelligence, an unexpected development considering their long history of disagreements over privacy, advertising and platform control.
Meta has introduced Muse Spark Open, a large open-weight AI model designed for advanced reasoning, software development and autonomous AI tasks. The company also collaborated with Apple to make the model run on Mac Studio computers using Apple’s MLX machine learning framework.
The partnership brings together Meta’s AI research and Apple’s hardware capabilities, giving developers another way to operate powerful AI systems without relying entirely on cloud infrastructure.
For Apple, the collaboration demonstrates the growing potential of its silicon architecture in advanced AI computing. For Meta, it creates another environment where developers can experiment with its models.
Meta Introduces Muse Spark Open With 816 Billion Parameters
Muse Spark Open represents a significant addition to Meta’s artificial intelligence portfolio. The model contains 816 billion total parameters, although only 42 billion become active during each processing step. It uses a mixture-of-experts architecture, allowing the system to handle demanding workloads without activating its entire network for every request.
The model also supports a context window of up to one million tokens, giving developers room to process lengthy documents, extensive software projects and complicated instructions. Meta developed Muse Spark Open for tasks that require sustained reasoning rather than simple chatbot conversations, including advanced coding, technical analysis and AI agents capable of completing multiple stages of work.
Meta is positioning the release as a competitive option for developers who want greater control over sophisticated AI systems. Its open-weight approach allows users to download and customize the model, although commercial use remains subject to licensing conditions.
Apple Mac Studio Provides the Hardware for Local AI Processing
Running an AI model with hundreds of billions of parameters typically requires substantial computing infrastructure. Meta and Apple have demonstrated that high-end desktop systems can provide another option, using two Mac Studio computers equipped with Apple’s M5 Ultra processors.
Each Mac Studio can support up to 512GB of unified memory, giving the combined configuration enough memory capacity to accommodate the model’s enormous requirements. Apple’s unified memory architecture allows processing components to access a shared memory pool, reducing some of the complications associated with moving data between separate memory systems.
The demonstration does not mean Muse Spark Open will run on every Mac. Its hardware requirements remain significant, and the two-machine configuration represents a considerable investment. Nevertheless, the ability to operate a model of this scale outside conventional AI server infrastructure highlights the expanding capabilities of Apple Silicon.
Apple’s MLX Framework Makes the Collaboration Possible
Apple’s MLX framework plays an important role in bringing Muse Spark Open to Mac Studio. The open-source machine learning framework was developed specifically for Apple Silicon, allowing researchers and developers to build and execute AI workloads while taking advantage of the company’s processor architecture.
Meta worked with Apple to optimize Muse Spark Open for MLX, improving compatibility with the Mac Studio configuration. The collaboration demonstrates how software optimization can make large AI models more practical on specialized hardware, particularly when developers want to operate systems locally.
Apple has increasingly emphasized the machine learning capabilities of its processors, and this project gives the company another example of those capabilities in action. Rather than focusing only on consumer AI features, the collaboration places Mac Studio within the broader discussion about advanced AI development and infrastructure.
Mark Zuckerberg and Apple Leadership Highlight an Unexpected Partnership
The collaboration attracted attention partly because Meta and Apple have not always maintained a friendly relationship. The companies have repeatedly disagreed over privacy protections, digital advertising and the influence of major technology platforms.
Meta CEO Mark Zuckerberg publicly acknowledged Apple’s contribution to the Muse Spark Open project, while Apple’s leadership also recognized the technical work behind the collaboration. Their public exchange reflected an unusual moment of cooperation between two companies that have frequently criticized each other’s business decisions.
Meta’s AI leadership also indicated that the companies had worked together for several months to make the model compatible with Apple’s hardware. The arrangement appears focused on a shared technical objective rather than a broader change in their competitive relationship.
For both companies, the benefits are practical. Meta gains another deployment platform for its AI technology, while Apple demonstrates that its computers can handle increasingly demanding machine learning workloads.
Muse Spark Open Expands Opportunities for Developers and AI Agents
Meta is making Muse Spark Open available through several major AI platforms, including Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Developers can also access the model through inference providers such as Groq, OpenRouter, Fireworks AI and Together AI, giving them different options depending on their computing resources and project requirements.
The company designed Muse Spark Open for agentic workflows, where AI systems perform sequences of tasks rather than simply generating responses to individual prompts. Meta demonstrated the model’s capabilities through Muse Code, using it to make software changes to Ice Cubes, an open-source Mastodon application. The company also tested the model with DaVinci Resolve, allowing the AI to examine video footage and identify suitable moments for vertical editing.
These demonstrations suggest that Muse Spark Open could support software development, video production and other complex workflows. Developers will still need to consider hardware costs, inference performance and the model’s licensing restrictions before adopting it for commercial applications.
Meta and Apple Partnership Highlights Growing Demand for Local AI
The Meta and Apple collaboration reflects growing interest in running advanced artificial intelligence systems on locally controlled hardware. Cloud platforms remain essential for many large-scale AI operations, but local deployment can offer advantages for organizations that want greater control over their information, software environments and computing resources.
Apple’s Mac Studio configuration demonstrates how high-memory desktop systems can support workloads traditionally associated with specialized AI servers. Meta benefits by making its model available across more hardware platforms, while Apple gains an opportunity to demonstrate the capabilities of its processors beyond consumer applications.
There are practical limitations. A configuration involving two M5 Ultra Mac Studio computers with substantial unified memory is expensive, and Apple’s 512GB configuration was not yet available for purchase when the model launched. Local deployment also does not guarantee better performance than cloud infrastructure, particularly for organizations processing large numbers of requests simultaneously.
What Comes Next for Muse Spark Open and Apple Silicon?
Muse Spark Open adds another major competitor to the expanding market for open-weight artificial intelligence. Its combination of large-scale reasoning, multimodal capabilities and local deployment options could attract developers interested in building advanced AI applications without depending exclusively on proprietary cloud services.
For Meta, the release supports its broader efforts to expand access to powerful AI models. For Apple, it strengthens the case for using Mac Studio and MLX as serious development tools for demanding machine learning projects. The collaboration could also encourage more AI developers to optimize their models for Apple Silicon, although neither company has confirmed plans for a wider partnership.
The most interesting development may be the relationship itself. Meta and Apple have spent years disagreeing over major technology policies, yet both companies found a reason to cooperate on advanced AI computing. Their work on Muse Spark Open shows how the demand for increasingly powerful AI systems is creating partnerships that once seemed unlikely.
Sources
- The Verge — Meta and Apple team up to make Muse AI models run better on Macs
- Meta AI Research — Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date
- TechCrunch — Meta launches an open-weight AI model that runs on Macs
- MacRumors — Meta and Apple Bring Muse Spark Open AI Model to Mac Studio
- Axios — Apple and Meta bury the hatchet — for A
