Apple and Meta Team Up to Run Muse Spark Open on Macs
Apple and Meta have collaborated to run the Muse Spark Open AI model on Mac Studio computers, marking a notable shift in their long-standing rivalry. According to Meta, the model has 816 billion total parameters with 42 billion active and a context window of one million tokens. It runs on two Mac Studios equipped with Apple’s M5 Ultra chip, using the updated MLX software framework. Alexandr Wang, Meta’s chief AI officer, said the model’s heavy memory demands make it a poor fit for most hardware but well suited to Macs due to their unified memory capacity. Each M5 Ultra chip supports up to 512GB of memory, allowing two linked machines to provide a terabyte of unified memory sufficient to hold the entire model. Muse Spark Open is available for download from Hugging Face and runs on inference engines including SGLang and vLLM. Apple updated MLX to split open-source models across multiple machines, enabling this deployment.
Apple Launches HomeView Powered by Siri AI
Apple has introduced HomeView, a smart home device powered by Siri AI and Apple Intelligence. The device features a 7-inch Liquid Retina display, facial recognition for personalized content access, and integration with Apple Intelligence to draw on personal context from iPhone apps like Messages, Calendar, and Reminders. HomeView supports FaceTime calls, music playback, home monitoring, and hands-free automations via voice control. It can be placed on a tabletop or wall-mounted, with pre-orders starting October 13, 2026, and availability beginning November 13, 2026. The device comes in silver, starlight, rose pink, and space gray finishes. According to Greg Joswiak, Apple’s senior vice president of Worldwide Marketing, HomeView is designed to be shared by everyone in the home while feeling personal to each user through facial recognition and adaptive interface adjustments based on user distance.
Technical Foundations: MLX Framework and On-Device AI
The collaboration relies on Apple’s MLX framework, which has been updated to efficiently distribute open-source AI models across multiple Mac machines. This enables large models like Muse Spark Open to run locally on-device, reducing reliance on cloud infrastructure. Meta reported testing Muse Spark Open on agentic workflows using its AI coding tool, Muse Code, including tasks such as shipping bug fixes, developing features for the Ice Cubes Mastodon client, and analyzing video footage in DaVinci Resolve frame by frame. The model is accessible via API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Its weights are hosted on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Meta emphasized that the model is natively multimodal and designed for coding, agentic work, and long-horizon reasoning tasks.
Implications for Apple’s AI Strategy and the Industry
Apple’s partnership with Meta signals a strategic openness to external AI innovation while advancing its own on-device AI capabilities through Apple Intelligence and Siri AI. By enabling powerful open-weight models to run on Mac hardware, Apple strengthens its position in the AI development ecosystem, particularly for developers requiring high memory and computational resources. The launch of HomeView extends Apple Intelligence into the home environment, combining privacy-focused design with contextual awareness. Together, these moves reflect Apple’s focus on integrating AI into professional workflows and everyday life through tightly integrated hardware, software, and services. The collaboration also highlights a growing trend of cross-industry cooperation in AI, even among historical rivals, to push the boundaries of what open models can achieve on consumer-grade hardware.
Muse Spark Open’s Role in Meta’s AI Ecosystem and Developer Tools
Meta’s release of Muse Spark Open extends beyond hardware compatibility, embedding the model into its broader AI development ecosystem. The model’s weights are hosted across multiple cloud platforms—including Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models—ensuring broad accessibility for developers. Meta explicitly designed Muse Spark Open for **agentic workflows**, testing it with tools like **Muse Code**, an AI-powered coding assistant. These tests included real-world applications: debugging and deploying fixes for the **Ice Cubes Mastodon client**, automating feature development, and even analyzing video footage in **DaVinci Resolve** by reviewing frames, identifying key moments, and applying vertical cropping adjustments. According to Meta, the model’s **native multimodal capabilities** and **long-horizon reasoning** make it uniquely suited for complex, multi-step tasks in software engineering and media production.
The collaboration with Apple also underscores Meta’s push to democratize access to high-performance AI tools. While Muse Spark Open’s **816 billion total parameters** and **1TB memory requirement** make it impractical for most consumer hardware, Meta emphasized its compatibility with **Mac Studio’s M5 Ultra chips**—which support up to **512GB of memory per chip**—allowing developers to link **two to four machines** for unified memory pooling. The model’s availability via API platforms like **Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI** further lowers barriers to entry, enabling developers to integrate it into existing workflows without heavy infrastructure investments. Meta’s chief AI officer, **Alexandr Wang**, framed the partnership as part of a broader vision: *“We think that a lot of the world will run on open models, and they should be American,”* signaling a strategic alignment with Apple’s on-device AI initiatives while reinforcing Meta’s commitment to open-source innovation.
Technical documentation released by Apple highlights how the updated **MLX framework** enables seamless distribution of open-source models across multiple Mac machines, a feature critical for Muse Spark Open’s deployment. The framework’s optimizations—including support for **SGLang and vLLM inference engines**—allow developers to run the model locally without cloud dependencies, aligning with Meta’s emphasis on **privacy-preserving, on-device AI**. Apple’s benchmarks and documentation further clarify that the collaboration leverages the M5 Ultra’s **unified memory architecture**, a key differentiator for handling models of this scale. By making Muse Spark Open accessible via **Hugging Face without extra configuration for Apple silicon**, Meta and Apple have created a low-friction pathway for developers to experiment with state-of-the-art AI tools on consumer-grade hardware, potentially accelerating innovation in fields like **automated software development, media analysis, and agentic systems**.
Sources
- Meta and Apple set aside years of rivalry to run Muse Spark Open AI model on Macs
- Apple
- Meet Apple HomeView, a new way to enjoy entertainment and manage the home
Photo: BoliviaInteligente / Unsplash (Unsplash License)
