Meta and Apple Unite to Run Muse Spark AI on Macs

Overview

Meta and Apple have announced a significant collaboration to bring high-performance artificial intelligence models to desktop Macs, marking a notable shift in their competitive landscape. The two tech giants unveiled that their new Muse Spark Open model is optimized to run efficiently on M5 Ultra Mac Studios using Apple’s MLX framework.

This partnership allows users to execute some of the world’s most capable open-weight AI models locally, rather than relying solely on cloud-based infrastructure. Both CEOs, Mark Zuckerberg and John Ternus, publicly endorsed the integration, highlighting the technical achievements and the mutual benefits for their respective ecosystems.

A Strategic Collaboration Between Rivals

Meta has long been a competitor to Apple in the consumer technology space, yet this launch represents a growing trend of cooperation in the hardware and software layers of AI. Meta Chief AI Officer Alexandr Wang commented on the relationship to Axios, noting that the companies have been collaborating for several months.

“We’re excited for mommy and daddy to be playing nice again,” Wang said, referencing the historical rivalry between the two Silicon Valley titans.

The joint effort ensures that Meta’s latest open-weights model can leverage Apple’s specialized silicon effectively. This synergy benefits both parties: Meta expands the reach of its open-source AI initiatives, while Apple reinforces the Mac’s position as a powerful station for local generative AI workloads.

Technical Specifications and Performance

Announced today, Muse Spark Open is described by Meta as the “most capable US open-weight model to date.” The architecture is a natively multimodal model featuring 816 billion total parameters. Despite its massive scale, the model is designed to handle complex tasks such as long-horizon operations, sophisticated software engineering, and multi-step reasoning.

Crucially, Meta claims this is the first model in its class capable of running on a desktop environment utilizing two Apple M5 Ultra Mac Studios. This performance is made possible through deep integration with the MLX framework, Apple’s array framework for machine learning tasks on Apple Silicon.

“Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple,” Meta CEO Mark Zuckerberg stated on X.

Apple’s new technical post elaborates on the capabilities of MLX, demonstrating how it handles Muse Spark Open alongside two other open models. The optimization allows the hardware to process these enormous parameter counts efficiently, maintaining performance levels that were previously only achievable on specialized server farms.

Leadership Endorsements and Market Impact

The public support from the highest levels of both companies underscores the strategic importance of this union. Apple CEO John Ternus responded directly to Zuckerberg’s announcement, expressing pride in the engineering team’s accomplishments.

“Congrats to the team!” Ternus wrote. “Amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!”

This endorsement serves as a strong signal to developers and enterprise customers who are increasingly seeking to deploy large language models (LLMs) locally for privacy and latency reasons. By enabling such a large model to run on dual M5 Ultra units, Apple is effectively raising the bar for what constitutes a viable desktop AI workstation.

Implications for the Open AI Ecosystem

Meta’s decision to release the weights of Muse Spark Open aligns with its broader strategy to lead the open-source AI movement. However, making such a model accessible on consumer hardware like the Mac Studio democratizes access to high-end AI capabilities.

Key takeaways from this development include:

• Decentralized AI: Users can now run billion-parameter models without constant reliance on cloud APIs, reducing latency and enhancing data privacy.
• Hardware Validation: The collaboration validates the M5 Ultra chip as a serious contender for professional AI research and development.
• Software Integration: The success of this launch highlights the maturity of the MLX framework in handling large-scale tensor operations on unified memory architectures.

Looking Ahead

As the release of Muse Spark Open weighs becomes available, the tech community will be watching closely to see how the model performs in real-world scenarios beyond benchmark tests. The collaboration between Meta and Apple suggests that future advancements in AI will increasingly depend on tight hardware-software co-design.

While Meta continues to innovate in model architecture, Apple’s focus remains on embedding these capabilities into the devices users already own. This partnership may well set a precedent for how major tech companies navigate the balance between competition and necessary collaboration in the rapidly evolving AI landscape.

For now, the message is clear: the era of requiring industrial-grade data centers to run state-of-the-art open AI models is coming to an end, at least for those with access to the latest Apple hardware.