Hacker News new | past | comments | ask | show | jobs | submit
drivebyhooting | on: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
Slower is meaningfully dumber when you’re time bounded and need all the inference time compute you can get.