GoKawiilMuse Spark Open model released with optimizations for Apple's M5 Ultra and NVIDIA/AMD GPUs
A new open-weight AI model, Muse Spark Open, has been released and optimized to run on Mac Studio machines with Apple's M5 Ultra chip, reportedly allowing two linked Macs to function like a single machine with a terabyte of unified memory. The release also includes custom Helion GPU kernels in SGLang that the developers say speed up performance on NVIDIA and AMD hardware, and the model underwent additional safety post-training evaluated against an internal scaling framework before release.
GoKawiil's interpretation of the reporting above, not reported fact.
By optimizing for consumer-grade Apple hardware alongside enterprise GPUs, the release could broaden access to powerful on-device AI for developers without large cloud budgets, letting them run agentic coding and media-editing workflows locally. The emphasis on safety evaluation suggests the developers are trying to address concerns specific to open-weight models, where anyone can modify or misuse the released weights once public.
- Muse Spark Open is optimized to run on Mac Studio with Apple's M5 Ultra chip using MLX.
- Two linked Mac Studios can reportedly act as one system with up to a terabyte of unified memory.
- The open-weight release includes GPU kernel optimizations for NVIDIA and AMD hardware and underwent additional safety testing before launch.
Apple Mac Studio M5 Ultra — This is the exact machine the article highlights for running Muse Spark Open locally, thanks to its massive unified memory and huge memory bandwidth. If you want to experiment with large on-device models and agentic workflows like those described, a Mac Studio is the purpose-built desktop to do it.
See Apple Mac Studio M5 Ultra on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: research.meta.ai — Meta Superintelligence Labs, 2026-10-15
Published there as: “Muse Spark Open”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.