Liquid AI unveils LFM3 family for enterprise on-device deployment

Liquid AI released the LFM3 family on October 27, comprising six models across language, vision, speech, and structured decision-making. The models share a 1.7-billion-parameter foundation, with specialized versions at roughly 2 billion parameters. Enterprises can fine-tune each model on a single GPU and deploy it on devices, private servers, or cloud infrastructure.

The lineup includes the LFM3-1.7B text model, LFM3-VL-2B for vision, LFM3-Audio-2B for speech, LFM3-Omni-2B for multimodal input, and the d1-2B and d1-Omni-2B decision models. Liquid said the models use a tri-hybrid architecture combining convolution and attention, pretrained on 22 trillion tokens.

In company-reported benchmarks, LFM3-1.7B scored 38.42 on WebArena-Verified and 41.29 on AutomationBench's public evaluation, outperforming Qwen3.5-9B in those tests. Liquid acknowledged that larger models remain stronger at agentic coding and knowledge tasks, and its documentation advises against using the model for complex multi-step reasoning.

Liquid listed API pricing of $0.08 per million input tokens and $0.20 per million output tokens for the text model, with cached input at $0.02. The LFM3-1.7B weights are released under the Liquid Open License v1.0, and companies with more than $10 million in annual revenue need a separate commercial agreement. Mercedes-Benz and Shopify were named as pre-release testing partners, but the announcement did not include independent validation of the benchmark results.