A new benchmark, Prophet Arena, reveals that AI models are forecasting real-world events with comparable accuracy to traditional prediction markets. Developed by researchers at the University of Chicago, this platform pits AI systems against live predictions from markets like Kalshi and Polymarket, assessing their abilities on unresolved events ranging from elections to sports outcomes. AI models demonstrated unique prediction styles, with GPT-5 leading in accuracy among them. Unlike historical data testing, Prophet Arena emphasizes future predictions, leveling the playing field for various AIs and allowing for a more systemic comparison of human judgment against machine forecasting. The highest returns were noted for OpenAI's o3-mini model, which emerged as the profit champion for its betting accuracy. The results suggest that AI could potentially redefine how institutions approach risk assessments and strategic decision-making by leveraging data without emotional bias.

Source 🔗