Since Artificial Intelligence began to become an increasingly common tool in people’s lives, the poker industry raised the question of whether one day these AI models will be good enough to consistently defeat humans. This, in turn, opened another question: which is the best AI model to play poker?
Since the question was posed, several competitions between language models have been played. The first AI match took place back in 2020. After nearly 4,000 hands, Meta’s model called LLAMA 4 ended up losing and OpenAI o3 was the winner. However, Artificial Intelligence advances so quickly that these results are no longer the same.
GTO Wizard developed its own model, called GTO Wizard AI. The advantage of this model is that it was created specifically for poker, as it is part of the agent that runs the solver itself. This technology was incorporated into the brand in 2023 and they have been perfecting it over the past years.
“Through intensive learning, GTO Wizard AI considers each particular situation as it occurs in the game and solves them in real time, within seconds,” GTO Wizard states.
Read more From photographer to champion: Katerina Lukina shone in Jacksonville — CodigoPoker
Finally, they put GTO Wizard AI to compete with other language models, which were not specifically designed for poker, and the difference in results has been huge. GTO Wizard AI recorded a winrate of 19.4BB/100 throughout a match. This is an extremely good mark, especially considering that a professional human achieves, on average, a winrate of 5BB/100.
GTO Wizard shared the achievement on their social media: “We outperformed all the main AI models in poker. GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro, Grok 4 and more. They all played 5,000 hands of heads-up no-limit against our poker agent. They all lost.”
The results of the other artificial intelligence models were drastically different. The different versions of GPT, OpenAI’s model, recorded winrates between -16 and -18.2. Claude, for its part, recorded a winrate of -20.4. The best performer was the Marvel model from MIT, with a result of -14. All these numbers were recorded after 5,000 hands played.
For now, the trend remains that conventional AI models fail a lot in strategic decision-making when playing a poker game. GTO Wizard AI, on the other hand, correctly solves most of the situations it faces, just as expected from a solver.
Read more Jungleman and Texas Mike clashed sparks in a six-figure pot during Hellmuth’s Home Game —
