Artificial-intelligence systems can forecast real-world events as accurately as crowd-based prediction markets, according to new research from the University of Chicago’s SIGMA Lab. The study introduces Prophet Arena, a benchmark that pits large language models against live markets on questions ranging from election outcomes to sports results and economic indicators. Early results show OpenAI’s GPT-5 topping the leaderboard with a Brier score of 82.21, while the smaller o3-mini model generated the highest simulated betting returns. In several cases the machines diverged sharply from market consensus yet still proved correct, suggesting AI can identify mispriced probabilities. The researchers say testing models on unresolved, forward-looking events eliminates the possibility of training-data leakage and offers a clearer measure of real-time reasoning. As platforms such as Kalshi and Polymarket begin integrating their own AI tools, the findings point to a future in which institutional decision-makers may weigh machine forecasts alongside, or even ahead of, the wisdom of crowds.