LiquidAI Releases LFM3-1.7B — a 1.7B-Parameter Model for Local Agents

On October 23, 2026, the model LiquidAI/LFM3-1.7B with the text-generation task was published on HuggingFace. The author positions it as a text model for local agents and specialized applications, built on a new tri-hybrid architecture. The model underwent post-training for agent harnesses Hermes Agent, Pi and OpenClaw, and its context window is 256K tokens. In the first days the page gained 6311 downloads and 212 likes.

What it means

The model weighs 1.7 billion parameters, and the largest weights file is 3.2 GB in safetensors format. It runs via transformers. There are already 20 quantized builds on the platform: GGUF, NVFP4, MLX and FP8. The smallest working build is Q4_0 at 1 GB, meaning the model fits on a phone or laptop. The license is the LFM Open License v1.0: commercial use is permitted up to annual revenue of 10 million US dollars, and the threshold does not apply to non-profit organizations. When distributing, you must include the license, note changes and retain attribution notices. The author explicitly states that the model is not recommended for agentic coding and complex multi-step reasoning due to its small size. For tasks requiring knowledge, they advise using retrieval or search. In the same line, LFM3-1.7B-Base and LFM3-Omni-2B were released in the same days. You can follow new models for local deployment in the CPU3D news feed.