Liquid AI's Compact Models Challenge Giants

Liquid AI has launched LFM3, a new family of compact AI models designed to enhance enterprise control over AI workloads by enabling operations on company-controlled hardware. The October 27 release includes six models built on a shared 1.7-billion-parameter foundation, with specialized versions around 2 billion parameters.

These models are particularly beneficial for companies automating repeatable workflows, such as request routing and information extraction. Liquid AI reports that its smallest text model can outperform larger, rival models in specific browser automation and software tool-use tasks. Enterprises can customize and deploy these models on devices, private servers, or cloud infrastructure.

The LFM3 family offers specialized models for language, vision, and speech, with the flagship LFM3-1.7B text model supporting a significant context window. While Liquid AI's benchmarks show strong performance in areas like browser automation and function-calling tests, the company acknowledges that larger models still excel in complex coding and reasoning tasks.

Liquid AI provides both downloadable models under an open license and a hosted service with specific pricing. For local deployments, companies with over $10 million in annual revenue require a separate commercial agreement. The business case for LFM3 depends on specific workloads, deployment costs, and licensing terms.