Liquid AI's D1 Vision Model Achieves 85-97% Quality Inspection Accuracy Without Task Training

Liquid AI has demonstrated that its D1 decision model, now equipped with vision capabilities, can classify manufacturing defects across diverse products with 85-97% accuracy without requiring task-specific training. The company tested D1 on an industrial inspection application reviewing parts from four production lines—circuit boards, candles, cashews, and chewing gum from the public VisA dataset—and found the model could accurately sort good and defective parts immediately, based solely on brief task descriptions.