Meta launches an open-weight AI model that runs on Macs
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Meta’s Muse Spark Open is an 816B-parameter, 1M-token multimodal open-weight model that can run locally on Macs, but it needs at least 1TB of RAM and is best suited to an M5 Ultra Mac Studio. For production teams, this makes private local coding and desktop-agent workflows more viable on Apple hardware, but it is not a laptop-class deployment target and will still require high-end workstation provisioning or API access.
Meta's new 816B-parameter open-weight model (1M context, natively multimodal, tuned for agentic coding) can run locally on a Mac—but only on an M5 Ultra Mac Studio with at least 1TB of RAM, so this is a workstation-class deployment, not laptop inference. If you can hit that hardware bar, you get a frontier-tier open model you can run on-prem with full weight access, cutting per-token API costs and keeping sensitive codebases and agent workflows entirely local.