Liquid AI releases open-weight LFM3 models
LFM3 is a 2B-class family on one backbone: text, vision, audio, omni, and two decision models, with open weights.
Official source (opens in a new tab)
Liquid AI's post on 27 October 2026 announces LFM3, a new generation of Liquid Foundation Models for what it calls specialized intelligence. The first release is in the 2B parameter range. Six models branch from LFM3-1.7B-Base: a text model, a vision model, an audio model, an omni model, and two decision models. Liquid says the base and the post-trained models are open-weight and available on Hugging Face, the Liquid Playground, and the Liquid AI API.
LFM3-1.7B is the text model, with a 256K-token context window, post-trained for agent harnesses including Hermes Agent, Pi, and OpenClaw. Liquid says it fits in about 1 GB at 4-bit precision, and that it uses 35% fewer parameters than LFM2.5-2.6B while outperforming that model. LFM3-1.7B-Base was pre-trained on 22T tokens. LFM3-VL-2B is the vision model. LFM3-Audio-2B takes text or audio and can answer with text, speech, or both, and the post says it supports 18 languages. LFM3-Omni-2B combines vision, audio, and language. The decision models are d1-2B, on the vision model, and d1-Omni-2B, on the omni model. Liquid says they answer a list of questions in one forward pass, each with a probability, instead of writing a reply. The post says each model can be fine-tuned on a single GPU.
Who it's for: someone who wants a small open-weight model to fine-tune and run on a device.
Fits: Best for Vibe Coding stays Cursor. Liquid says larger models still win on agentic coding and knowledge. Best Agent Builder is the system around a model, and LFM3 is a weight family. The verdicts are unchanged.