LFM2 2.6B

LFM2-2.6B is a highly efficient 2.6 billion-parameter language model from Liquid AI designed to run locally on edge devices like laptops and phones while delivering performance comparable to much larger models. Trained with pure reinforcement learning and featuring a hybrid architecture combining convolutions and attention, it outperforms models three times its size on instruction-following and reasoning benchmarks without the cost of cloud infrastructure.

Available for Chat, Vision, and File Uploads.