The competitions to see which AI is better at playing games are already underway
Google has kicked off a new AI showdown that puts poker in the spotlight, this time as part of a broader test of how leading language models handle games built on strategy, psychology, and risk. The competition is taking place inside Google’s Game Arena, where several top LLMs are being ranked against each other.
The event features three games with very different demands. Chess is used to measure structured reasoning, while Werewolf focuses on deception and social reading. Poker sits in the middle, blending logic with uncertainty and pressure. The poker matches are played heads-up, highlighting decision-making without full information and forcing the models to balance aggression and caution.
Well-known figures from both poker and chess are involved in presenting the matches. Doug Polk, Liv Boeree, and Nick Schulman are handling poker coverage, joined by chess grandmaster Hikaru Nakamura. The first round aired earlier this week, with two more poker sessions scheduled over the next two days during morning livestreams.
What sets this contest apart is its scale. A similar AI poker experiment last year used only a few thousand hands, which many felt was too small to draw real conclusions. This time, the models will play roughly 900,000 hands, offering a much deeper look at long-term tendencies, risk tolerance, and adaptation.
The results are likely to spark mixed feelings within the poker community. While AI-versus-AI games can be fascinating from a technical angle, they also underline growing concerns about bot use in real games. Strong performances by LLMs may impress researchers, but for many players, they also serve as a reminder of how important game integrity and detection tools have become.
Google’s experiment shows how poker continues to be a proving ground for artificial intelligence, even as the stakes around its real-world impact keep rising.
