Liquid AI Releases LFM3: Six Open-Weight 2B-Class Multimodal Models

Liquid AI has launched LFM3, a family of six open-weight multimodal models in the 2B parameter class, designed for specialized deployment across text, vision, and other modalities. The models are positioned for developers who need capable multimodal inference at a scale suitable for edge or resource-constrained environments, without depending on closed APIs. LFM3 builds on Liquid AI's non-transformer architecture research, which has previously demonstrated competitive performance at smaller parameter counts. Developers can access the models via Hugging Face, making fine-tuning and integration into existing ML pipelines straightforward. This release is notable for expanding the open-weight multimodal landscape beyond the dominant transformer paradigm.