Meta Muse Spark Open Runs Locally on Mac Studios
Meta's new open model runs locally on Mac Studios, enabling powerful AI inference without cloud dependency.
Local Inference on Mac Hardware
How It Stacks Up Globally
Strategic Shift for Meta AI
Meta has launched Muse Spark Open, an 816B-parameter multimodal model designed for coding and agentic tasks. In a notable partnership with Apple, Meta optimized the model to run locally on Mac Studios using the MLX toolkit. This setup allows two M5 Ultra Mac Studios to function as an AI supercomputer, handling tasks like iOS app fixes and video editing without cloud reliance.
On Artificial Analysis’ Intelligence Index, Spark Open scores 39, placing it ahead of most Western open-weight models but behind Chinese competitors like GLM 5.3 and Qwen3.8. While it matches rivals on multimodal and knowledge tasks, Chinese models currently edge it out on coding performance. This highlights a persistent gap in Western open-source development compared to rapid advancements in China.
This release signals Meta's strategic pivot, balancing its closed frontier model Wonder with open weights for developers. It empowers engineers to run serious AI on existing hardware, reducing latency and costs. However, for tasks requiring top-tier coding capabilities, the trade-offs remain significant against international rivals.
A powerful step for local AI development, though Chinese models still lead in coding.