Liquid AI's LFM3 packs text, vision, audio, and decision models into a 2B-parameter family that runs on-device
Introducing LFM3: Every Modality for Specialized Intelligence
Liquid AI's LFM3 family puts text, vision, audio, omni, and decision models on one shared backbone, all around 2B parameters and open-weight. Trained on 22T tokens with a tri-hybrid mix of gated convolutions, sliding-window attention, and global attention, LFM3-1.7B matches larger models on agentic tasks and decodes 1.8–2.8× faster than rivals on phones, laptops, and a Raspberry Pi 5, with a 256K context window.
Per parameter, LFM3-1.7B matches LFM2-1.2B on short prompts across edge devices, and with 8K tokens in context it is about 25% faster.