RobustMAD: Evaluating Real-World Robustness of Multimodal Small Language Models for Deployable Anomaly Detection Assistants

Small vision-language models beat GPT-5 Nano at spotting factory defects, but all of them still hallucinate too often to trust on the line.