OpenAI’s o3 model took first place in a five-day poker tournament featuring nine leading language models, ultimately winning $36,691. The event, organized by PokerBattle.ai, tested how AI handles uncertainty, adaptation, and strategic thinking. O3 won by consistently playing according to theory. Claude from Anthropic and Grok from X.ai followed with significant wins as well.
Most models performed well but struggled with bluffing, positioning, and overly aggressive play. Some, such as Meta’s Llama, were eliminated early on. The showdown demonstrated AI’s improved decision-making under pressure but also revealed persistent weaknesses that reflect the challenges of real-world decision-making.
