Meta has unveiled what the company describes as its most capable open-weight artificial intelligence model to date, released for users in the United States. Known as Muse Spark Open, the model carries a notable distinction beyond its raw performance figures - it has been built from the ground up to operate natively on Apple Mac hardware.

The model operates on an architecture that draws on 816 billion parameters and supports a context window of up to one million tokens. According to evaluation data published by Meta, Muse Spark Open outperforms other leading open-weight models currently available in the market. The company released these benchmark results on Thursday alongside the model's debut.

Built for Complex Tasks and Local Deployment

Muse Spark Open is natively multimodal, meaning it can process and reason across different types of input without requiring additional adaptation layers. The model has been purpose-built for demanding use cases, including sophisticated coding assignments and agentic workflows - tasks where an AI system operates with a degree of autonomy to complete multi-step objectives.

One of the more technically ambitious aspects of the release is its support for local execution on Apple computers. Running a model of this scale locally is not a trivial undertaking, however. According to reports, users will need a minimum of one terabyte of RAM to operate it on their machines. For optimal performance, the recommended hardware is the Mac Studio configured with Apple's M5 Ultra chip, which provides the memory bandwidth and unified memory architecture necessary to handle a model of this complexity.

Despite the steep hardware requirements, the potential payoff for users who can meet those thresholds is considerable. Meta has positioned Muse Spark Open as a serious productivity tool for coding and other desktop-based professional tasks.

Real-World Agentic Deployments

To demonstrate the model's practical capabilities, Meta described several real-world deployments it conducted internally. In one instance, the company paired Muse Spark Open with Muse Code, its dedicated AI coding terminal, and ran the combination through a series of agentic workflows. Among the tasks completed was a comprehensive coding project involving a Mastodon client, which encompassed bug fixes, the introduction of new features, and a complete build of the application.

In a separate demonstration, Meta deployed the model in conjunction with DaVinci Resolve, the professional-grade video editing software developed by Blackmagic Design. The model proved capable of reviewing video footage and performing cropping operations within that environment, illustrating its ability to extend beyond text-based tasks into more visually oriented workflows.

The Case for Open-Weight AI

The release of Muse Spark Open sits within a broader industry conversation about the relative merits of open-weight versus closed, proprietary AI models. Open-weight models offer several structural advantages that make them attractive to developers, researchers, and enterprises alike.

• They tend to be significantly more cost-efficient to deploy, since users are not locked into subscription or per-query pricing structures imposed by proprietary providers.
• They offer a level of transparency that closed-source models cannot match, allowing developers to inspect model behavior and better understand outputs.
• They provide flexibility for customization and fine-tuning, enabling organizations to adapt the model to specific use cases without seeking permission from a central vendor.

These characteristics have driven growing adoption of open-weight models across the technology sector, particularly among teams that prioritize data privacy, cost control, or the ability to run models in isolated or offline environments.

Availability and Access

Meta has made Muse Spark Open available immediately through a range of third-party API platforms, giving developers multiple access points through which to integrate the model into their own applications and services. The company has not restricted distribution to a single channel, which aligns with the open-weight philosophy of broad and accessible deployment.

The release marks another significant step in Meta's ongoing effort to establish itself as a leading contributor to the open AI ecosystem, competing in a space that includes other notable open-weight releases from organizations across the industry. Whether the hardware requirements will limit adoption among individual users remains to be seen, but for well-resourced teams and professionals with access to high-end Apple hardware, Muse Spark Open represents a substantial new option in the landscape of locally deployable AI models.