Liquid AI's Small Model Beats Larger Ones in Data Extraction

Liquid AI has released its smallest language model yet, the LFM2.5-230M, and it is already turning heads. Despite having only 230 million parameters, the model outperforms competitors that are four times its size in data extraction tasks, proving that bigger is not always better in the world of AI. The model is designed to run efficiently on devices with limited computational resources, including smartphones, laptops, and even robotics hardware. This makes it ideal for edge computing scenarios where cloud connectivity is unreliable or where latency must be minimized. Data extraction—the ability to pull structured information from unstructured text—is a critical task for enterprises. From parsing invoices and contracts to extracting medical records, accurate extraction can save countless hours of manual labor. Liquid AI’s model achieves state-of-the-art results on several benchmark datasets, matching or exceeding the performance of models with billions of parameters. “We focused on efficiency without compromise,” said a Liquid AI spokesperson. “Our goal was to create a model that delivers enterprise-grade performance in a compact form. The LFM2.5-230M proves that you don’t need a massive server farm to get accurate results.” The model’s small size also means lower energy consumption, which is increasingly important as companies seek to reduce their carbon footprint. It can run entirely on-device, ensuring data privacy since sensitive information never leaves the user’s device. Liquid AI has made the model available for developers and enterprise customers, with APIs and SDKs for popular programming languages. Early adopters report that the model integrates easily into existing workflows and provides reliable results even on older hardware. The release of the LFM2.5-230M is part of a broader trend toward smaller, more specialized AI models. While large language models like GPT-4 dominate the headlines, many real-world applications benefit from models that are fast, cheap, and private. Liquid AI’s achievement suggests that the future of AI may not be about ever-larger models, but about smarter, more efficient designs.