Liquid AI introduces LFM3, six small models across text, vision, audio and omni

Liquid AI announced LFM3, a new generation of its Liquid Foundation Models built on what it calls a new tri-hybrid architecture, with new multimodal encoders. It says the architecture is optimized for 14 processor classes across GPUs, CPUs and NPUs, and that the models can be fine-tuned on one GPU.

The family has six models:

• LFM3-1.7B: text, 256k context window.
• LFM3-VL-2B: vision-language, with video input.
• LFM3-Audio-2B: tool calls, voice chat and reasoning in 18 languages.
• LFM3-Omni-2B: combines vision, audio and language.
• d1-2B and d1-Omni-2B: decision models with typed outputs and calibrated probabilities in a single pass; the Omni version adds audio input.

Liquid's own claims: LFM3-VL-2B performs competitively with Gemma 4 12B (six times larger), Gemini 3.5 Flash Lite and GPT-5.6 Luna, and LFM3-1.7B is first on the Pareto frontier of device intelligence on its Pipette tool. These are not independently verified.

Weights are on Hugging Face (the 1.7B model's card lists the LFM1.0 license), and Liquid offers a playground and an API. The announcement had about 5,000 points on X by October 28.