Skip to content
Play Poker Online Now! — technology — Spin to Win! — Spin the Wheel of Fortune

AI Poker Showdown: Who Won the $100K No-Limit Texas Hold'em Battle?

No humans allowed—just AI bluffing, betting, and busting in a high-stakes poker experiment. Who outsmarted the rest, and who went broke?

,   and  Administrator
2 min read
The picture is inside a club, there are two snooker tables, on the first table two people are...
The picture is inside a club, there are two snooker tables, on the first table two people are standing the first person is playing the game, in the background many people are standing and talking behind them there is a cream color wall, to the wall there are few photo posters.

AI Poker Showdown: Who Won the $100K No-Limit Texas Hold'em Battle?

In October 2025, a unique online poker tournament took place—one without human players. Instead, eight AI models competed in a $10/$20 no-limit Texas Hold'em game, each starting with a $100,000 bankroll. The event, organised by recreational player and IT specialist Max Pavlov, tested how well large language models could handle the complexities of multi-player poker.

The game ran across four tables simultaneously, with each AI bringing a distinct playing style. OpenAI emerged as the strongest performer, finishing with a profit of $36,691. Its strategy closely followed game-theory-optimal (GTO) principles in opening pots but strayed by 3-betting too often and folding less than ideal.

Gemini took the most aggressive approach, playing many hands and frequently making large 3-bets. At the opposite end, DeepSeek played the tightest, raising preflop half as often as GTO suggests and folding far less to 3-bets. Magistral, meanwhile, was the most cautious before the flop but turned highly aggressive afterwards, continuation-betting 88% of the time and rarely folding to pressure.

Claude and Grok adopted similar strategies, seeing fewer flops than GTO would recommend and failing to fold enough against continuation bets. The weakest performer was Meta’s LLAMA, which went broke after 3,501 hands.

The event echoed past AI successes in poker, such as Libratus, developed by Carnegie Mellon University. In 2017, Libratus defeated top human professionals—Daniel McAreevy, Jason Les, and Dong Kim—in a 20-day heads-up no-limit Hold'em challenge in Pittsburgh, winning by 1.8 big blinds per 100 hands. However, its dominance did not extend to full-ring multi-player games, where the complexity of bluffing and modelling multiple opponents proved far greater.

The tournament highlighted both the strengths and limitations of AI in poker. OpenAI’s model secured the biggest win, while others struggled with balance—either playing too loosely or too conservatively. The results suggest that even advanced AI still faces challenges in adapting to the unpredictable nature of multi-player poker.

Read also:

Related

Latest