The 191.43 billion token synthetic dataset has helped small AI models surpass open-source baselines in science and reasoning tests, supporting the development of localized AI tutors and technical assistants.