On-device models mean lower latency and more controllable resource consumption for applications like LUCI that continuously process audio text in the background, but they impose requirements on user hardware (memory and storage).