{"version":1,"type":"story","url":"https://digestai.news/story/meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs","json":"https://digestai.news/story/meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs.json","markdown":"https://digestai.news/story/meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs.md","slug":"meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs","headline":"Apple benchmarks Muse Spark Open on Mac Studio with MLX and RDMA","summary":"Apple’s Machine Learning Research team demonstrated how its MLX framework enables running large language models locally on Mac Studio hardware. The team benchmarked three frontier models—Muse Spark Open (816B parameters), Mistral Medium 3.5 (128B), and Qwen 3.8 (27B)—on both M3 Ultra and M5 Ultra Mac Studio configurations. Muse Spark Open, a mixture-of-experts model, requires distributed inference across multiple Mac Studios due to its size, but MLX’s support for RDMA over Thunderbolt 5 and the M5 Ultra’s GPU Neural Accelerators improves performance by up to 4.6× in time-to-first-token compared to the M3 Ultra.\n\nThe research highlights that Qwen 3.8 and Mistral Medium 3.5 fit on a single Mac Studio with 512GB of RAM, while Muse Spark Open needs parallelization. The team also showed how MLX’s `mlx.launch` and `mlxlm.server` tools enable distributed inference and agentic AI workflows locally, with examples using OpenCode. The benchmarks emphasize Apple silicon’s efficiency for LLM inference, with power consumption as low as 200W for M3 Ultra and 300W for M5 Ultra.","keyPoints":["MLX on Mac Studio runs Muse Spark Open (816B params) via distributed inference across multiple Macs","M5 Ultra’s GPU Neural Accelerators cut time-to-first-token by up to 4.6× vs. M3 Ultra","MLX supports RDMA over Thunderbolt 5 for low-latency multi-Mac communication"],"whyItMatters":"This shows Apple silicon can compete with cloud for local LLM inference, enabling developers to run large models privately without high costs or latency.","category":{"slug":"hardware","name":"Hardware & Compute","url":"https://digestai.news/category/hardware"},"entities":{"companies":["Apple"],"models":["Muse Spark Open","Mistral Medium 3.5","Qwen 3.8"],"people":[]},"firstPublishedAt":"2026-10-14T21:02:00Z","updatedAt":"2026-10-15T18:15:36Z","sourceCount":8,"hasPrimarySource":true,"sources":[{"outlet":"Apple Machine Learning Research","title":"Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio","url":"https://machinelearning.apple.com/research/mlx-distributed-inference","publishedAt":"2026-10-15T00:00:00Z","type":"primary","primary":true,"lead":true},{"outlet":"The Verge AI","title":"Meta and Apple team up to make Muse AI models run better on Macs","url":"https://theverge.com/news/1012293/meta-apple-muse-spark-open-ai-model-mlx","publishedAt":"2026-10-15T18:15:36Z","type":"press","primary":false,"lead":false},{"outlet":"Unite.AI","title":"Meta’s 816B-Parameter Muse Spark Open Weights Are Now Public","url":"https://unite.ai/metas-816b-parameter-muse-spark-open-weights-are-now-public","publishedAt":"2026-10-15T17:35:16Z","type":"press","primary":false,"lead":false},{"outlet":"cnbc.com","title":"Meta says it worked with Apple to make Muse Spark Open run locally on M5 Ultra Mac Studios, letting users pair two to four Studios to code or do agentic work","url":"https://cnbc.com/2026/10/15/apple-meta-partner-ai.html","publishedAt":"2026-10-15T17:20:27Z","type":"press","primary":false,"lead":false},{"outlet":"research.meta.ai","title":"Meta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window","url":"https://research.meta.ai/blog/muse-spark-open","publishedAt":"2026-10-15T17:10:01Z","type":"press","primary":false,"lead":false},{"outlet":"cryptobriefing.com","title":"Meta’s Muse Spark Open AI model can run locally on Apple Mac Studio","url":"https://cryptobriefing.com/apple-meta-muse-spark-open-mac","publishedAt":"2026-10-15T10:08:00Z","type":"press","primary":false,"lead":false},{"outlet":"Hugging Face","title":"Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta","url":"https://huggingface.co/blog/meta-muse-spark-open","publishedAt":"2026-10-15T00:00:00Z","type":"primary","primary":true,"lead":false},{"outlet":"engadget.com","title":"Meta built a huge open-weight AI model for Macs, but you'll need 1TB of RAM to run it","url":"https://engadget.com/2287833/meta-built-a-huge-open-weight-ai-model-for-macs-but-youll-need-1tb-of-ram-to-run-it","publishedAt":"2026-10-14T21:02:00Z","type":"press","primary":false,"lead":false}],"sourceNotes":null,"discussions":[{"site":"Hacker News","url":"https://news.ycombinator.com/item?id=50106620","points":31}],"thread":{"title":"Muse Spark Open Expands Across Tech Giants","url":"https://digestai.news/thread/apple-benchmarks-muse-spark-open-on-mac-studio-with-mlx-and-rdma","storyCount":2},"cite":{"text":"Digest AI, \"Apple benchmarks Muse Spark Open on Mac Studio with MLX and RDMA\", 14 October 2026, https://digestai.news/story/meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs","publisher":"Digest AI","title":"Apple benchmarks Muse Spark Open on Mac Studio with MLX and RDMA","datePublished":"2026-10-14T21:02:00Z","url":"https://digestai.news/story/meta-releases-open-weight-muse-spark-model-for-apple-silicon-macs"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}