Search
Articles · 3
Meta launches Muse Ember, its smallest AI model for agent subtasks
Meta has released Muse Ember, which it calls its smallest and most efficient AI model for agent subtasks such as tool calls and coding. It is available as muse-ember-1.0 at $0.10 per million input tokens.
Meta details Prometheus, its Ohio AI supercluster headed for about 1 million GPUs
Meta says the first phase of Prometheus, its gigawatt-scale AI supercluster in Ohio, is complete and trained its Muse models. In an engineering post, Meta said the site will scale to roughly 1 million GPUs.
Meta unveils Muse Spark Open, 816B-parameter model for M5 Ultra Mac Studio
Meta has unveiled Muse Spark Open, an 816B-parameter open-weight multimodal model optimized for the M5 Ultra Mac Studio. The model supports on-device agentic coding and reasoning workflows through Apple’s MLX framework.
From the feed · 8
Xavier Niel called French lawmakers incompetent ignoramuses over a new per-gigabyte tax. The amendment passed the Assembly 11 to 10.
AMD announced that its EPYC processors are seeing success at Meta. The post references an earlier note from @Muse.
Researchers enumerated 6,800 Muse share links via passive DNS records. The links exposed private details including loan applications, executive notes, and revenue figures, prompting Meta to suspend the sharing feature.
6,800 Muse conversation-sharing links turned up via a single DNS query by Claroc researchers. Among the hundreds examined were a loan request, a bank fraud alert, and an entire company's vacation schedule.
OnlyFans creators say Meta restricts their Instagram accounts for adult content while allowing its Muse AI agent to sext users in the adults-only app. One performer with over 100,000 followers called the policy unfair.
Meta released Muse Spark Open, an open-weight model it optimized for Apple’s Mac Studio with M5 Ultra using Apple’s MLX framework, 9to5Mac reports. Meta AI chief Alexandr Wang told Axios the two companies have been working together for several months.
1.5 teraflops from 30 Mac Minis powers a Cray-1 replica at Spain's computer history museum. The setup uses 70 cores across five quad-core i7s and 25 dual-core i5s, beating the 1976 original's 160 megaflops peak.
Meta says its new Muse Wonder and Muse Wonder Full AI models have caught it up with the frontier. Both add "active memory" and "active focus" features meant to help agents work through long tasks.