A new study reveals that leading AI systems from Google, OpenAI, and Anthropic lost money betting on Premier League matches during the 2023-24 season. The research, conducted by General Reasoning, tested eight top AI models in a virtual betting environment using historical data and team statistics.

Anthropic's Claude Opus 4.6 performed best with an average 11% loss, nearly breaking even on one attempt. Google's Gemini 3.1 Pro managed a 34% profit on one try but went bankrupt on another. xAI's Grok 4.20 went bankrupt once and failed to complete the other attempts. All models lost money over the season, with many experiencing complete financial ruin.

The study highlights AI's limitations in analyzing real-world scenarios over extended periods, contrasting sharply with AI's recent successes in software engineering. General Reasoning's CEO Ross Taylor emphasized that many AI benchmarks fail to reflect real-world complexity, offering reassurance to professionals concerned about AI replacing human jobs in various industries.