Hacker News new | past | comments | ask | show | jobs | submit
petu | on: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
Datacenter hardware can batch at large scale, probably over 90% more energy efficient per token than a MacBook.