Liquid AI

MIT CSAIL spinout (2023) behind the LFM (Liquid Foundation Model) line of small, edge-targeted models, released under its own lfm1.0 licence. Founded by ramin-hasani (CEO), daniela-rus, Alexander Amini and Mathias Lechner.

Reached first in this corpus through maxime-labonne's lfm2-5-230m-chess, a third party's chess fine-tune of LFM2.5-230M-Base — evidence about what someone did with a Liquid model, not about the model line itself. lfm2-technical-report (T1, Nov 2025) supplies that first-party account: LFM2 is a hybrid architecture (gated short convolutions + a few grouped-query-attention blocks) found by hardware-in-the-loop search, spanning 350M–8.3B parameters, pretrained on 10–12T tokens, open-weight on Hugging Face.

Related

ramin-hasani · daniela-rus · maxime-labonne · lfm2-5-230m-chess · lfm2-technical-report