started · updated
Liquid AI launches LFM3 family of small, open-weight models
Liquid AI has released the LFM3 family, a collection of six open-weight multimodal models designed for local deployment on hardware such as laptops, phones, and single GPUs. Unlike many industry trends toward massive parameter counts, these models are approximately 2 billion parameters each and share a common foundation called LFM3-1.7B-Base.
The lineup includes specialized variants for text, vision-language, audio, decision-making, and an “omni” model capable of processing text, images, video, and audio simultaneously. The text model features a 256K-token context window, while the vision-language model, LFM3-VL-2B, reportedly matches the performance of models six times its size on vision tasks.
Technically, the models utilize a tri-hybrid architecture combining gated short convolutions, sliding-window attention, and global attention. The models are available via Hugging Face and are optimized for various runtimes, including GGUF for llama.cpp and MLX for Apple Silicon. The small footprint allows the text model at 4-bit precision to occupy approximately 1 GB of memory.