- OpenAI’s o3 model won a five-day poker tournament with nine AI chatbots
- The O3 model won by playing the most consistent game
- Most of the top language models did well in poker, but struggled with bluffing, positioning, and basic math.
In an unprecedented digital collision, nine of the world’s most influential language models played high-stakes poker for five days.
OpenAI’s O3 and Anthropic’s Claude Sonnet 4.5 each fund.
When OpenAI’s o3 model came out $36,691 richer in a week-long poker game, there was no trophy, just bragging rights.
The experimental PokerBattle.ai was completely controlled by artificial intelligence and all players received the same initial message. It was pure strategy, if you can call it micro-decisions, made by machines that don’t really understand how to win or lose, or how humiliating it is to go bankrupt with seven deuces.
This was told in an unusual way for a technical gimmick. The most powerful AIs didn’t just bluff and play: they adapted, shaped their opponents, and learned to deal with ambiguity in real time. Although they did not play poker perfectly, they came very close to judging experienced players.
OpenAI’s O3 quickly proved to have the most stable hand, winning three of the five largest pots and sticking to textbook preflop theory. Anthropic’s Claude and X.com’s Grok round out the top three with notable earnings of $33,641 and $28,796 respectively.
Meanwhile, Llama lost his entire stack and was eliminated early. The rest of the field finished somewhere in the middle, with Google’s Gemini earning a modest profit and Moonshot’s Kimi K2 taking a chip hit of $86,030.
AI game
Poker has long been one of the best analogues for testing artificial general intelligence. Unlike chess or Go, which rely on perfect information, poker requires players to think under conditions of uncertainty. It is a reflection of real-world decision-making, from business negotiations to military strategy and chatbot development.
What always stood out about the tournament was that the bots were often too aggressive. Most people prefer intensive action strategies, even in situations where it would have been wiser to walk away. They tried more to win big pots than to avoid losing. And they were terrible at bluffing, not because they didn’t try, but because their bluffing was often due to misinterpreted hands rather than clever deception.
However, AI tools are getting smarter and go far beyond superficial intelligence. They don’t just repeat what they read; You make probability assessments under pressure and learn to read the room. It’s also a reminder that even the most powerful models still have flaws. Misunderstanding situations, jumping to conclusions and forgetting your “position” is not just a poker problem.
In a real poker room, you may never sit in front of a language model, but you will probably communicate with it and try to make important decisions. This game was just a taste of what could be.
