{"id":42874,"date":"2026-10-18T08:36:23","date_gmt":"2026-10-18T06:36:23","guid":{"rendered":"https:\/\/www.mixtv1.com\/index.php\/2026\/10\/18\/apple-ran-metas-816b-muse-spark-open-on-four-mac-studios-at-over-36-tokens-a-second\/"},"modified":"2026-10-18T08:36:23","modified_gmt":"2026-10-18T06:36:23","slug":"apple-ran-metas-816b-muse-spark-open-on-four-mac-studios-at-over-36-tokens-a-second","status":"publish","type":"post","link":"https:\/\/www.mixtv1.com\/index.php\/2026\/10\/18\/apple-ran-metas-816b-muse-spark-open-on-four-mac-studios-at-over-36-tokens-a-second\/","title":{"rendered":"Apple ran Meta&#8217;s 816B Muse Spark Open on four Mac Studios at over 36 tokens a second"},"content":{"rendered":"<\/p>\n<p>Apple has run Muse Spark Open, an 816-billion-parameter model from Meta with 42 billion parameters active at a time, across four Mac Studios and measured over 36 tokens a second. <a href=\"https:\/\/machinelearning.apple.com\/research\/mlx-distributed-inference\" target=\"_blank\" rel=\"noopener\">Apple\u2019s machine learning researchers say<\/a> the model &#8220;cannot fit on a single machine&#8221;, and their figures put four M5 Ultra machines splitting it by width at 36.73 tokens a second at an 8,192-token context, against 15.74 for two machines in a pipeline.<\/p>\n<\/p>\n<p id=\"caption-attachment-36767\">The banner Apple published with the research post on October 15, alongside the MLX release the benchmarks were run with. Image: Apple<\/p>\n<p>Memory is why it takes four. Muse Spark Open &#8220;would require more than 1.6TB of RAM in its native precision (BF16) for the model parameters alone&#8221;,<\/p>\n<p> <a class=\"echo_read_more\" href=\"https:\/\/mixed-news.com\/en\/apple-mlx-muse-spark-open-four-mac-studios-36-tokens\/\" target=\"_blank\">\u00a0\u00bb\u00a0More Info >>><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Apple has run Muse Spark Open, an 816-billion-parameter model from Meta with 42 billion parameters active at a time, across four Mac Studios and measured over 36 tokens a second. Apple\u2019s machine learning researchers say the model &#8220;cannot fit on a single machine&#8221;, and their figures put four M5 Ultra machines splitting it by width<\/p>\n","protected":false},"author":55,"featured_media":42875,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[339],"tags":[348,6072,333,101,344,36,6020],"class_list":["post-42874","post","type-post","status-publish","format-standard","has-post-thumbnail","category-vr","tag-ai","tag-ai-research","tag-apple","tag-artificial-intelligence","tag-meta","tag-mixtv","tag-pc-hardware"],"_links":{"self":[{"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/posts\/42874","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/users\/55"}],"replies":[{"embeddable":true,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/comments?post=42874"}],"version-history":[{"count":0,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/posts\/42874\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/media\/42875"}],"wp:attachment":[{"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/media?parent=42874"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/categories?post=42874"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.mixtv1.com\/index.php\/wp-json\/wp\/v2\/tags?post=42874"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}