Liquid AI releases LFM3 open-weight model family

Liquid AI announced LFM3, the next generation of its Liquid Foundation Models. The first release has six models, all branched from a single LFM3-1.7B-Base backbone. Four cover input modalities (text, vision, audio and omni), and two are "decision" models specialized from the vision and omni variants. Base and post-trained weights are available on Hugging Face, Liquid's Playground and the Liquid AI API.

The architecture came out of the company's hardware-in-the-loop search, which now spans 14 device and backend combinations. That is up from one phone and one laptop for LFM2. The new set includes Samsung Galaxy S26 Ultra and OnePlus 15 phones, a Raspberry Pi 5, Apple M4 Max and AMD Ryzen AI laptops, cloud CPUs, GPUs and Qualcomm's Hexagon NPU. Of the model's 24 layers, 13 are gated short convolutions, 5 use sliding-window attention with a 256-token window, and 6 use global attention. Liquid says LFM3-1.7B matches LFM2-1.2B per parameter on short prompts and runs about 25% faster with 8K tokens in context.

The base model was pre-trained on 22T tokens. The final 3.5T were a mid-training stage weighted toward agentic traces and tool calls, which also extended the context window to 256K tokens.

The text model, LFM3-1.7B, is post-trained for agent harnesses such as Hermes Agent and OpenClaw. In Liquid's own benchmarks it scores 45.60 on LiveClaw, 41.29 on AutomationBench and 43.40 on SWE-bench Verified. On several agentic tests it beats Qwen3.5 4B and Gemma 4 E4B, and on some it beats Qwen3.5 9B. Liquid says the model can be adapted with a short fine-tuning run on a single GPU. These results have not been independently verified.