Liquid AI releases LFM3 open weights for local multimodal workloads

Liquid AI released six open-weight LFM3 models on Hugging Face for immediate download and fine-tuning on a single consumer GPU, Tech Times reported. The family covers text, vision, audio and structured decisions, with a shared 1.7 billion-parameter backbone. The text model supports a 256K-token context window and fits in approximately 1 GB at 4-bit precision. Its architecture combines short convolutions, attention over a limited token window and global attention to reduce the memory used to retain prior context. The vision model adds video input, while the omni model accepts text, images, video and audio and produces text and speech. Liquid AI's reported performance gains have not been independently reproduced.