Humans outperform AI at this highly rigorous mathematics test

A new mathematics benchmark, published in Nature on June 12, 2026, demonstrates that current artificial intelligence systems still lag behind elite human mathematicians in solving novel and complex problems. The benchmark, designed to test AI on problems it has not been trained on, revealed significant gaps in AI's reasoning and problem-solving capabilities when faced with unfamiliar mathematical challenges. While AI models have shown impressive progress in areas like pattern recognition and data analysis, this rigorous test highlights their limitations in genuine mathematical insight and creative problem-solving. The study's findings suggest that achieving human-level mathematical intelligence requires more than just advanced algorithms and vast datasets; it necessitates a deeper understanding of mathematical principles and the ability to generalize knowledge to entirely new contexts. Researchers involved in the study emphasized that while AI can assist mathematicians by automating certain tasks, it cannot yet replicate the intuitive leaps and abstract reasoning that characterize human mathematical breakthroughs. The benchmark is expected to guide future AI development, pushing researchers to focus on improving AI's capacity for abstract thought and genuine understanding rather than solely on performance on existing problem sets.