LFM3 is a family of six open-weight models of about 2B parameters (learned settings) each, covering text, vision, audio, an all-in-one omni model and two decision models.
Liquid AI built six small models, like a toolbox, that you can fine-tune on a single GPU. They cover text, vision, audio, an omni model for all three, and two decision models that return probabilities.
Liquid says the 1.9B vision model rivals Gemma 4 12B, six times larger, and the 1.7B text model fits in about 1 GB. Weights are open, with day-0 support from Arm and others.
Key Capabilities:
- Six models: LFM3-1.7B (text), VL-2B (vision), Audio-2B, Omni-2B, plus d1-2B and d1-Omni-2B decision models.
- Long text memory: LFM3-1.7B reads up to 256K tokens (word chunks).
- Runs on your device: Fits about 1 GB at 4-bit, runs in LM Studio, and fine-tunes on one GPU.