LFM3-1.7B
LFM3-1.7B is a compact text model developed by Liquid AI, featuring 1.73 billion parameters and designed specifically for local agents, tool calling, and specialized workflows. Released in October 2026, it serves as part of the LFM3 model family, aiming to deliver efficient intelligence on edge devices while supporting an expansive 256K-token context window.
The model is built on a tri-hybrid architecture consisting of 24 layers. This design combines gated short-convolution layers, sliding-window attention layers, and global attention layers. This hybrid configuration enables the model to process long inputs efficiently while maintaining high performance on structured reasoning and instruction-following tasks.
LFM3-1.7B is post-trained for agentic workflows and structured outputs, making it well-suited for applications involving function calling, browser interactions, and automation harnesses. Despite its small parameter footprint—requiring roughly 1 GB of memory under 4-bit quantization—the model provides a resource-efficient option for developers seeking local execution capabilities.
Create with Crafiq
Generate images, 3D models, video and audio in one studio.
Explore the studioHow LFM3-1.7B ranks
LFM3-1.7B is highlighted in the table below. Switch the metric to see how the ordering changes.