Liquid AI Ships a 350-Million-Parameter Model That Runs Agentic Loops on Edge Devices
LFM2.5-350M is small enough to fit in under 500MB quantized but handles tool use and data extraction — the capabilities where small models typically fail hardest.
Liquid AI released LFM2.5-350M on Tuesday, a model purpose-built for agentic workflows that runs at just 350 million parameters, as announced by @liquidai. When quantized, the model fits in under 500MB — small enough to run on mobile devices, embedded systems, and other constrained environments where calling out to a cloud API introduces unacceptable latency or connectivity risk.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.