GoKawiilLiquid AI launches LFM3 multimodal model family at 2B parameter scale
Liquid AI released LFM3, a new generation of Liquid Foundation Models spanning text, vision, audio, omni, and decision variants built on a shared backbone. The initial release includes six 2B-class models trained from a base model called LFM3-1.7B, using a new hybrid architecture and 22 trillion tokens of pre-training. The base and post-trained weights are open and available now on Hugging Face, Liquid AI's Playground, and its API.
GoKawiil's interpretation of the reporting above, not reported fact.
Liquid AI says LFM3-1.7B outperforms its earlier LFM2.5-2.6B model while using 35% fewer parameters, suggesting efficiency gains that could matter for running capable models on local devices rather than cloud servers. By supporting fine-tuning on a single GPU and fitting in roughly 1 GB at 4-bit precision, the models appear aimed at developers wanting specialized, edge-deployable AI rather than general-purpose giants. This positions Liquid AI to compete in the growing market for small, efficient, customizable models rather than chasing scale.
- LFM3 unifies text, vision, audio, omni, and decision models on one shared 2B-parameter backbone.
- Liquid AI claims LFM3-1.7B beats its prior LFM2.5-2.6B model despite using 35% fewer parameters.
- Weights are open and available on Hugging Face, Liquid AI's API, and Playground, with support for local fine-tuning on a single GPU.
Source: liquid.ai, 2026-10-27
Published there as: “Introducing LFM3: Every Modality for Specialized Intelligence”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.