Liquid AI
Liquid AI is a private foundation-model company founded in 2023 by researchers from MIT CSAIL. Its current LFM models target local deployment across text, vision, audio and combined inputs.
Company & leadership
- Co-founder and CEO
- Ramin Hasani
- Co-founder and CTO
- Mathias Lechner
Founders
- Ramin Hasani, Mathias Lechner, Alexander Amini and Daniela Rus
Research origins and company launch
Liquid AI grew from work at MIT's Computer Science and Artificial Intelligence Laboratory. The company emerged from stealth in December 2023 and announced $46.6 million in seed funding led by OSS Capital and PagsGroup. Its four founders combined research in machine learning, recurrent models and robotics.
The people leading Liquid
Ramin Hasani leads Liquid AI as CEO and Mathias Lechner as CTO. Co-founder Daniela Rus is also the director of MIT CSAIL. The current team page identifies Jeffrey Li as chief operating officer, responsible for execution across product, engineering and research.
Commercial work beyond the model releases
In June 2025, Liquid and G42 announced a partnership to develop and deploy private generative AI systems. The agreement included model development and infrastructure work with G42 companies such as Core42 and Inception. This partnership does not change Liquid's US headquarters or make G42 the author of every LFM checkpoint.
Coverage and access
Practical guidance. These articles cover the current LFM3 and LFM2.5 generation families, Liquid Nanos, encoders, retrieval models and d1 decision models. Deprecated releases remain visible for migration context. Historical checkpoints, base-only variants and precision conversions are being reviewed separately. Hosted API limits and downloadable checkpoint limits are stated separately.
A named license with commercial conditions
LFM Open License v1.0 includes a US$10 million annual-revenue threshold for commercial use and a defined nonprofit exception. The license is not Apache 2.0. Review the terms attached to the exact checkpoint and the relevant organization before deployment.
Hosted limits and local limits differ
Liquid's API catalog currently lists five generation models, each with a 128,000-token window. Some checkpoint cards publish different limits and input formats. The model pages identify these separately; one deployment's context size should not be transferred to another.