Liquid AI Debuts Six Open-Weight Models for Local AI Workloads

Liquid AI announced its LFM3 family on Oct. 27, introducing six open-weight models designed to run and be fine-tuned on local hardware. Each has roughly two billion parameters, and the company says fine-tuning can be done on a single GPU.

The lineup covers text, vision-language, audio and decision-making tasks. It also includes an omni model that handles text, images, video and audio.

The six models build on the LFM3-1.7B-Base foundation. The text model supports a context window of 256,000 tokens, the amount of text it can process at once.

The architecture combines gated short convolutions, sliding-window attention and global attention. The shared base was pre-trained on 22 trillion tokens.

Liquid AI is positioning the models for uses including robotics. Their local deployment and fine-tuning design lets developers adapt the models for specific applications and run them on their own hardware.

LFM3-VL-2B matches models more than six times its size on vision tasks. Those results do not establish comparable performance across vision tasks generally.