Meta has unveiled Muse Spark Open, an open-weight AI model with 816 billion parameters, engineered to operate natively on Macs equipped with Apple silicon. The model leverages a massive context window and multimodal capabilities, making it suitable for complex coding and video workflows.
- Muse Spark Open uses 816 billion parameters and a 1M token context window.
- Designed to run locally on Macs with at least 1TB RAM and M5 Ultra chip.
- Supports complex tasks like coding and video editing workflows.
What happened
Meta has launched Muse Spark Open, a highly capable open-weight AI model that runs natively on Macs using Apple silicon. This release marks a significant collaboration that tailors AI technology specifically for Apple's hardware environment. The model incorporates an enormous 816 billion parameters and supports a 1 million token context window, enabling it to process complex and lengthy inputs effectively.
Muse Spark Open is multimodal and geared towards agentic tasks such as coding and video editing. Meta demonstrated its capabilities by integrating it with Muse Code, their AI coding terminal, and applying it to real-world workflows, including improving a Mastodon client with bug fixes and feature additions, as well as working with DaVinci Resolve to assist with video footage review and cropping.
Why it matters
The launch of Muse Spark Open highlights the growing trend of powerful AI models being optimized for local desktop use rather than relying solely on cloud infrastructure. Running such a large model locally provides users with enhanced privacy, responsiveness, and control over their AI workflows. This is particularly valuable for professionals in software development and media production who seek real-time assistance without cloud dependency.
Meta’s open-weight approach also introduces cost efficiencies and transparency benefits compared to proprietary closed models. Developers and enterprises gain flexibility to customize and deploy AI more openly, fostering innovation and potentially reducing vendor lock-in. The model’s multimodal and agentic capabilities position it as a versatile tool for next-generation AI applications on consumer hardware.
What to watch next
Adoption rates of Muse Spark Open will depend heavily on user access to the required high memory capacity and Apple’s latest silicon, particularly the Mac Studio with its M5 Ultra chip. It will be important to observe how Meta and partners expand accessibility or optimize the model for devices with lower specs to broaden its audience.
The integration of Muse Spark Open with varied workflows, including coding environments and video editing software, suggests a roadmap toward deeper AI-assisted productivity tools. Monitoring Meta’s API deployments and ecosystem partnerships will reveal whether this open-weight model can spur wider innovation and challenge established AI platforms both on Macs and other systems.