QVAC, a Tether subsidiary, has released the QVAC Genesis III dataset comprising 191.43 billion tokens designed to train smaller AI models for STEM reasoning. The dataset targets enhanced performance in science, technology, engineering, and mathematics applications through specialized training data. The release covers 19 STEM disciplines aligned with established curriculum systems across high school, university, and professional education levels. This structured approach aims to improve AI model accuracy and reasoning capabilities within technical and scientific domains.