Liquid AI Releases LFM3, Six Open-Weight Models for On-Device AI

Liquid AI released LFM3, a new generation of its foundation models that puts text, vision, audio, omni and decision models on one shared backbone, the company said in a blog post. The first release covers six open-weight models in the 2 billion parameter class, available on Hugging Face, in Liquid AI's playground and through its API.

The models are trained from LFM3-1.7B-Base, which was pretrained on 22 trillion tokens. Liquid AI says LFM3-1.7B outperforms its earlier LFM2.5-2.6B with 35% fewer parameters. The text model fits in about 1 GB at 4-bit precision.

The architecture keeps gated short convolutions for more than half of its layers and adds sliding-window attention. Liquid AI tuned it on 14 device and backend combinations, including Samsung and OnePlus phones, a Raspberry Pi 5, Apple and AMD laptops and Qualcomm's Hexagon NPU.

Each model can be fine-tuned on a single GPU. Liquid AI says the family is meant as a starting point for specialized models that run on the device where an application is deployed.