Liquid AI

Liquid AI builds Liquid Foundation Models (LFMs), foundation models developed from first principles rather than by extending existing transformer architectures. The models are designed as an alternative to traditional transformers, with an emphasis on efficiency and adaptability across deployment targets, from data centres to edge devices. Technical work spans foundation models, efficient AI/ML, adaptive neural networks and on-device inference. The company originated as a spin-off from MIT CSAIL. Its team is composed of engineers, researchers and business leaders. Working practices are described as highly autonomous: individuals operate on the critical path within their domains, ideas are judged on merit rather than hierarchy, and problems are expected to be raised early and fixed directly. The approach to research mirrors the approach to organisation - assumptions are torn down and rebuilt from scientific fundamentals rather than taken from existing playbooks. Internally, the company draws a parallel between its white-box models and a white-box organisation, with transparency treated as central to collaboration. Staff are expected to work across boundaries where needed, and stated priorities include alignment between company goals and individual growth, alongside team members' wellbeing.