Liquid AI's LFM3-1.7B is an open-weight agent model it says fits in about 1 GB

Liquid AI says LFM3-1.7B, the text model in its new LFM3 family, takes about 1 GB of memory at 4-bit, so it fits on any phone or laptop. It has 1.73B parameters and a 256K-token context window, and it's built for local agents and tool calling. It's one of six ~2B LFM3 models covering text, vision, audio, omni and decision tasks. The catch: the weights are open under the LFM Open License v1.0, but it's not open source. Commercial use is allowed only for entities under $10M in annual revenue. In our test on a CPU-only Linux box with no GPU, it downloaded and answered, but it got "what is a GGUF file?" wrong. Liquid's own model card says to use retrieval or search for knowledge-heavy tasks.

llama-cli -hf LiquidAI/LFM3-1.7B-GGUF:Q4_K_M