Meta released Muse Spark Open on October 15, making its weights available for download alongside access through hosted inference providers.
The natively multimodal model has 816 billion total parameters, with 42 billion active per token, and a one-million-token context window. Meta positions it for coding, multistep reasoning and extended agent tasks.
Desktop deployment comes with substantial hardware
Meta says the model can run through Apple’s MLX framework on two M5 Ultra Mac Studios, each supporting up to 512 GB of unified memory.
That configuration puts a large model on a desktop, but requires far more memory than an ordinary personal computer. Active parameter counts describe computation per token; they do not represent the full weight-storage requirement.
Support also extends to several inference frameworks and hardware platforms, giving developers alternatives to the Apple configuration.
Key benchmark
Check the license before building
The weights use the Muse Model License, which includes limited commercial terms. Businesses should review those conditions before incorporating the model into products.
Meta’s announcement expands deployment choices. Whether a particular setup is worthwhile will depend on workload performance, hardware expense and operating requirements.
Sources: https://research.meta.ai/