OpenAI Won The Artificial Intelligence Poker Battle – CodigoPoker

OpenAI Won The Artificial Intelligence Poker Battle - CodigoPoker
OpenAI won the artificial intelligence poker battle

This week, a new poker tournament between LLMs (large language models), commonly known as artificial intelligence models, took place. The main AI model brands put their machines to compete, and after three days of play, there was a clear winner: OpenAI.

The tournament was divided into three phases. In the initial one, something similar to quarterfinals was played, in which eight AI models faced each other in 1-on-1 duels. o3 beat Deepseek 3.2, Grok 4 lost to Gemini 3 Flash, GPT 5.2 defeated Gemini 3 Pro, and Opus 4.5 beat Sonnet 4.5. In the semifinals, o3 beat Gemini 3 Flash, and GPT 5.2 beat Opus 4.5. This is how o3 and GPT 5.2, both OpenAI models, reached the final.

Doug Polk 

United States
, who followed the event on his YouTube channel, commented: “I find it interesting that both models are from OpenAI. It seems we have the best poker-playing models here. And they both play similar styles, being hyper-aggressive and looking to exploit any weakness.”

After observing the performance of the different models, the YouTuber was able to draw some conclusions about their play and decision-making process. For example, he was struck by the effort it took them to properly identify suited draws.

Read also: The APL Championship Returns To GGPoker With US$36M

“It’s a common thing I notice in AIs,” Polk said. “If I had to point out their biggest leaks, the suited draw issue is certainly one of them. But it also seems they struggle to understand that folding represents 0 EV. All plays in poker are neutral compared to past plays. You don’t think about the chips you would lose by folding; in fact, they are already in the pot. You can only make decisions for the chips you have at that moment, so they fail in that reasoning,” the professional player concluded.

Poker AI

Finally, he offered a final reflection on the overall performance of all the models.

The three hyper-aggressive AIs ended up winning, which is interesting. Those that stayed in the middle were the models with conservative styles. I thought Opus and Sonnet played reasonably. They made a reasonable number of raises, defended a reasonable number of times. But it seems these things were not built to withstand the hyper-aggression of the AIs that kept attacking. Overall, a fascinating challenge,” Polk concluded.

Read also: The BSOP San Pablo Has a Confirmed Schedule With 35 Events

Translated from

Leave a Reply

Your email address will not be published. Required fields are marked *