Mostly True
Liquid AI has launched d1-3B, a multimodal decision-making model that, according to the company, surpasses models ten times larger in weight class. The compact model is designed for agentic and decision-oriented tasks rather than general chat.
Liquid AI positions d1-3B as a small multimodal model focused on decision-making tasks, where the company says it outperforms substantially larger models. Details of the benchmark comparisons, model architecture, and availability were not fully specified in the available information, and the claim of beating models ten times its size remains based on the company's own characterization.
The release comes amid intensifying global competition in efficient AI models, a space closely watched by Korean semiconductor and AI industry players for its implications for edge deployment and compute demand.
Liquid AI's claim of outperforming models at ten times its scale is broadly supported by available information, though independent benchmark confirmation was not established — the assessment is Mostly True.