A recent study by the AI startup General Reasoning has shown that even advanced AI models struggled to make profitable bets on soccer matches over a Premier League season. The “KellyBench” report tested eight top AI systems and found them unable to consistently turn a profit, with some going bankrupt multiple times.
The loss-making performances of Google’s Gemini 3.1 Pro and Anthropic’s Claude Opus 4.6 highlight the gap between AI's strengths in specific tasks like coding, versus its struggles in more complex, dynamic environments such as sports betting. Even xAI Grok 4.20 failed to complete two out of three attempts, suggesting that real-world unpredictability poses a significant challenge.
This outcome underscores the limitations of current AI models and their inability to fully grasp and adapt to evolving situations over extended periods. It raises questions about how far we can rely on AI for decision-making in unpredictable scenarios.
