Liquid AI is a foundation model company that spun out of MIT CSAIL. Its central work is the development of Liquid Foundation Models - a family of AI models designed from first principles, built to move beyond the transformer architectures that have dominated the field. The models are engineered to run across the full deployment spectrum, from large-scale data centers to resource-constrained edge devices.
The company's technical scope spans foundation model research, systems engineering, model efficiency, and edge deployment. A stated design goal is interpretability: Liquid AI describes its approach as building "white-box" models - architectures that are transparent and legible, rather than opaque. This philosophy, the company says, also extends to how the organisation itself operates.
Liquid AI brings together engineers, researchers, and business leaders. Its working culture is structured around high-trust autonomy - domain experts are expected to own and execute within their areas with limited oversight. The organisation places emphasis on meritocracy, with ideas evaluated on their merits rather than by the seniority of who proposes them. Role flexibility and early problem escalation are also explicitly part of how the team operates.