All frontier models, (o1, o1-pro, QVQ, gemini-flash-thinking) score exactly 0% on main questions of this benchmark.