Meta Releases Muse Spark Open: An 816B Parameter Multimodal Model Built for Macs
Meta has unveiled Muse Spark Open, an 816B parameter open-weight model with a 1M token context window capable of running on high-end Macs with 1TB of RAM.
Meta has introduced Muse Spark Open, a massive new 816-billion-parameter open-weight artificial intelligence model. Marketed by the company as the most capable open-weight offering in the United States, the system features a 1 million token context window and native multimodal capabilities. Beyond its sheer scale, a notable aspect of the release is its targeted compatibility with Apple hardware, enabling local execution on supported Mac systems.
According to benchmark evaluations published by Meta, Muse Spark Open demonstrates performance superior to other leading open-weight models across various tasks. The architecture was specifically developed to manage complex coding challenges and agentic workflows. By distributing the model weights openly, Meta is emphasizing the advantages of open-weight systems, which provide greater transparency, structural flexibility, and potential cost efficiency relative to proprietary, closed-source models.
Hardware demands and local Mac execution
Although Meta engineered the model to run locally on macOS, doing so requires exceptional hardware capacity. The system demands a minimum of 1 terabyte of RAM to execute locally, according to reporting from Engadget. Because of this footprint, Meta noted that the model operates best on a Mac Studio equipped with an M5 Ultra chip.
For users and organizations with access to such high-specification Apple hardware, this allows an 816-billion-parameter foundation model to function entirely on a local machine without routing queries through remote data centers. For environments unable to meet the 1TB memory requirement, Meta is also distributing Muse Spark Open through several API platforms.
Coding automation and desktop agent workloads
Meta evaluated Muse Spark Open's practical execution across complex multi-step workflows, centering on software engineering and desktop control. The model was paired with Meta's AI coding terminal, Muse Code, and tasked with autonomous agent operations. During these trials, the system managed software maintenance on a Mastodon client, successfully completing a scope of work that included bug fixes, new features, and a full build.
Meta also highlighted the model's multimodal abilities by integrating it with DaVinci Resolve, the professional video editing application. When applied to the software, Muse Spark Open demonstrated the capability to review video material and execute automated footage cropping. These implementations highlight Meta's focus on enabling AI models to interact directly with professional desktop tools rather than operating solely through text-based chat interfaces.
What it means for developers
The introduction of Muse Spark Open represents a notable step forward for developers seeking powerful open-weight alternatives to closed systems. Open-weight designs grant engineers deeper visibility into model behavior, enhanced data privacy, and the flexibility to deploy workloads across varied infrastructure. The 1 million token context window also allows engineering teams to feed entire repositories, detailed documentation, or lengthy multimodal inputs directly into a single prompt session.
At the same time, the hardware barrier for local deployment remains substantial. Requiring 1 terabyte of memory restricts local deployment to high-end workstations such as an M5 Ultra Mac Studio. For many development teams, purchasing this caliber of desktop hardware is impractical, making cloud-hosted API access the primary method for deploying Muse Spark Open in production.
As development teams weigh open-weight offerings against existing proprietary solutions, testing and benchmarking across multiple providers becomes a core requirement. Developers looking to evaluate different model families can try top AI models cheaply through one API at https://apixoai.online, making it straightforward to compare capabilities across leading systems without juggling multiple vendor accounts.
Overall, Muse Spark Open signals Meta's ongoing investment in the open AI ecosystem. Whether deployed on specialized local Mac hardware for autonomous tasks or queried through cloud endpoints, the model provides developers with a versatile new option for advanced coding and multimodal desktop automation.
Source: Meta launches an open-weight AI model that runs on Macs — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
