Meta Launches Muse Spark Open AI Model Designed for Macs
Meta released Muse Spark Open, a powerful open-weight AI model built with Apple to run locally on high-end Mac Studio desktop computers.
Meta released its new Muse Spark Open artificial intelligence model on Thursday, partnering with Apple to enable the software to run directly on desktop computers. The launch marks a rare technical collaboration between the two technology rivals. Chief executives Mark Zuckerberg of Meta and John Ternus of Apple publicly praised the joint effort to optimize the model for Mac hardware.
What is Meta's Muse Spark Open model?
Muse Spark Open is an open-weight AI model, which means developers can inspect, modify, and customize its underlying software. Meta described it as the most capable American open-weight model to date. The model contains 816 billion parameters, which are the internal settings that dictate how an AI processes data and makes decisions. It also features a 1 million token context window, allowing it to read and analyze massive amounts of information in a single query.
The software is natively multimodal, meaning it natively understands and processes different formats like text and visual media. Meta specifically built the model for complex computer programming, multi-step reasoning, and agentic labor, where the AI executes multi-step workflows autonomously.
How does Muse Spark Open run on Mac computers?
Meta collaborated with Apple over several months to ensure the model runs efficiently on desktop hardware. It relies on Apple's MLX framework, a software library designed for machine learning on Apple Silicon chips. Meta chief AI officer Alexandr Wang confirmed the close technical cooperation between the two organizations.
Running the model locally requires immense hardware resources. Users need a system with at least 1 terabyte of memory, or RAM, to operate the model on-device. Meta noted that the model is designed to run on two Apple Mac Studio computers equipped with M5 Ultra chips. For organizations without that level of desktop hardware, Meta is also providing access to Muse Spark Open through multiple cloud application programming interfaces (APIs).
What can the new AI model do?
Meta designed the model to handle demanding desktop applications and software engineering tasks. In internal tests, Meta paired Muse Spark Open with its Muse Code coding terminal to execute software updates. The AI successfully generated bug fixes, added new features, and completed full builds for a Mastodon application.
Meta also demonstrated the model's ability to assist with high-end creative work. When integrated with the video editing software DaVinci Resolve, Muse Spark Open successfully reviewed raw video footage and performed automated cropping.
Why open-weight models matter for business
Open-weight models offer key commercial advantages over closed, proprietary AI services. They give businesses greater transparency and control over their tools, allowing engineering teams to fine-tune models for custom tasks without sending sensitive data to third-party providers. Open-weight solutions can also reduce long-term infrastructure costs.
By proving that an 816-billion-parameter model can run on desktop computers using Apple's MLX software, Meta and Apple are expanding options for companies that want high-performance AI capabilities hosted directly on their own hardware.
Comments 0
No comments yet. Start the conversation.