Liquid AI releases LFM3, six open-weight 2B-class multimodal models

LFM3 covers text, vision, audio, omni and decision tasks. LFM3-1.7B uses a tri-hybrid backbone with a 256K context window and fits in about 1 GB at 4-bit quantization.

Why it matters: A long-context multimodal family that fits in roughly 1 GB targets phones and edge devices, where non-transformer hybrid architectures compete on memory efficiency. It adds to the recent wave of small local models for voice and agents, such as Nvidia's Nemotron 3.5 Bolt.