Meta has released Muse Spark Open, an open-weight, natively multimodal AI model featuring 816 billion parameters. Meta claims the release is the most capable U.S. open-weight model to date, designed to handle long-horizon planning, complex software engineering tasks, and multi-step reasoning. Unusually for the rival tech giants, Meta collaborated directly with Apple to ensure the model runs efficiently on desktop hardware.
According to Meta, Muse Spark Open is the first model in its parameter class capable of running on a desktop setup using two Apple M5 Ultra Mac Studios paired with Apple’s MLX framework. Both Meta CEO Mark Zuckerberg and Apple CEO John Ternus publicly praised the cross-company collaboration, while Meta Chief AI Officer Alexandr Wang noted that technical teams worked together over several months to optimize performance.
Alongside the model release, Apple published technical documentation regarding its MLX array framework for Apple Silicon. The collaboration highlights a growing push toward local, high-parameter inference on workstation-class desktop hardware, reducing reliance on centralized cloud data centers for developer workflows.
Why it matters
Demonstrates rare strategic alignment between Meta and Apple to establish Mac Silicon as a premier desktop platform for open models.
Proves technical feasibility of running ultra-large (800B+ parameter) models locally on specialized workstation hardware.
Strengthens the developer ecosystem around Apple’s open-source MLX framework for high-performance machine learning.
Source: theverge.com



