LFM2.5-230M

Liquid AI's smallest model (230M params, LFM2 architecture), built to run fast on CPUs/NPUs/GPUs for on-device agentic tasks on phones, robots, and home/network automation devices. Pre-trained on 19T tokens with a 32K context extension and distilled from LFM2.5-350M; competes with models more than twice its size on instruction following, data extraction, and tool use.