Liquid AI has unveiled its smallest model yet, and the numbers are impressive. The LFM 2.5-230M, with just 230 million parameters, is outperforming models four times its size at data extraction tasks — all while being compact enough to run “anywhere.”
This is a significant milestone in the ongoing trend toward smaller, more efficient AI models. While much of the industry has focused on scaling up, Liquid AI is proving that clever architecture and training can deliver outsized results from a fraction of the parameters.
The LFM 2.5-230M is designed for on-device deployment, meaning it can run on smartphones, edge devices, and even in browsers — opening up possibilities for privacy-preserving AI applications that don’t require cloud connectivity.
Data extraction, the model’s specialty, is a critical capability for businesses processing large volumes of unstructured text — think invoice processing, document analysis, and knowledge base population.
Liquid AI hasn’t disclosed exact benchmark numbers yet, but early results suggest this tiny model could have a big impact on how AI is deployed at the edge.
