Meta releases 816B Muse Spark Open model for Apple Silicon desktops

Meta AI Research released the Muse Spark Open model today, positioning it as the most capable open-weight artificial intelligence system developed in the United States. The release brings advanced agentic capabilities to desktop environments, specifically targeting high-performance hardware configurations for local deployment.

The model contains 816 billion total parameters, but only 42 billion are active during inference. It supports a context window of one million tokens, allowing for extensive data processing in single sessions. According to Techmeme, Meta AI Research announced that the weights are available today for immediate use.

Technical Specifications and Hardware Requirements

The Muse Spark Open model is designed for deployment on Apple M5 Ultra Mac Studios. As Threat Times reported on 2026-10-15, Meta previously published weights for its largest open-weight AI model, enabling multimodal agentic tasks on dual Mac Studios with 512GB memory each. This new release continues that trajectory of providing powerful local AI tools for Apple Silicon hardware.

Meta states that the model brings leading agentic capabilities to desktops for the first time on this specific hardware class. The combination of a large context window and active parameter efficiency allows for complex, multi-step reasoning tasks on Apple M5 Ultra Mac Studios.

Strategic Context and Model Availability

This release aligns with Meta's stated strategy regarding open-weight models. As Threat Times reported on 2026-10-15, Meta publishes coding-focused AI models while stating it will not open-source its most powerful systems due to safety concerns. Muse Spark Open represents a significant step in their open-weight portfolio, distinct from their closed frontier models.

The company emphasizes that this model is the most capable US open-weight option currently available. By releasing the weights today, Meta allows developers and researchers to inspect and deploy the model locally. This approach contrasts with the restricted access typical of their most advanced proprietary systems.

What should you do?

Security and technology professionals should evaluate the resource requirements for deploying large local models. Organizations considering local AI deployment must ensure their hardware meets the specified Apple M5 Ultra Mac Studio requirements. Reviewing the model's agentic capabilities is essential for understanding potential operational impacts.

What to do and how to stay safe: Meta AI Research

• Verify that your local hardware meets the Apple M5 Ultra Mac Studio specifications before attempting deployment.
• Review the model's agentic capabilities to understand how it interacts with local systems and data.
• Monitor Meta AI Research for additional documentation on safe deployment practices and parameter configurations.
• Assess your organization's data privacy policies regarding local AI processing versus cloud-based inference.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is the total parameter count of the Muse Spark Open model?

The Muse Spark Open model has 816 billion total parameters, with 42 billion active parameters during operation.

Is this model available for cloud-based deployment?

The source material specifies that the model brings capabilities to desktops on Apple M5 Ultra Mac Studios, indicating a focus on local deployment.

Does Meta consider this its most powerful AI model?

Meta states it is the most capable US open-weight model, but it has previously indicated it will not open-source its most powerful systems due to safety concerns.