LFMs, short for Liquid Foundation Models, are Liquid AI's family of small, efficient open-weight models made to run on phones, laptops, cars and other devices.
In plain words
The family covers text, vision and audio models, plus tiny Nano models trained for one task. The LFM2.5 generation ranges from 230 million to 24 billion parameters. Its largest member, LFM2.5-24B-A2B, is a reasoning model that uses only 2 billion parameters at a time, which keeps it quick enough to run private agents on your own machine. People and companies can use the models for free, including commercially, until their company earns more than $10 million a year.
A quick example
Imagine a travel app that plans your day and books tickets while your phone is offline. A small model running on the phone itself can read your request, call the app's tools and answer without sending anything to a server.
Where you'll meet it
- Assistants and agents that run on phones, laptops and cars.
- Apps that call Liquid AI's hosted API instead of running the model themselves.
Related
Sources
- DocsLiquid Foundation Models, Liquid AI · read 13 Oct 2026
- DocsLFM2.5-24B-A2B model card, Liquid AI on Hugging Face · read 13 Oct 2026
- DocsRun fast, private agents on-device with LFM2.5-24B-A2B, Liquid AI (Hugging Face blog) · read 13 Oct 2026
Checked by WhatIsAI against the sources above · 13 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.