ChatGPT defeats Grok 4 to win chess tournament for AI

About an hour before the final, OpenAI announced the launch of its 11th generation LLM, called GPT-5. However, the model ChatGPT used in the final was still o3 - the company's strongest reasoning product. Facing xAI's Grok 4, which had performed well in the group stage, o3 showed superior strength with an average move accuracy rate of 90.8%, compared to Grok 4's 80.2%, according to Chess.com.

In all four games, ChatGPT checkmated its opponent after 35, 30, 28 and 54 moves. World number two Hikaru Nakamura said Grok 4 seemed nervous and anxious in this match, so it did not perform as well as in the previous two. It often lost pieces easily in the final match, which rarely happened in victories over Gemini 2.5 Flash and Gemini 2.5 Pro from Google.

The o3 model ended the tournament with three wins with a score of 4-0, with an average accuracy rate of 12 games above 91%. Although its strength is not comparable to that of a grandmaster, players with an online chess rating of 2,000 or lower may have difficulty facing o3, especially in blitz or super blitz games.

ChatGPT, a product of OpenAI, is a pioneer chatbot in the AI revolution, launched on Nov. 30, 2022. At that time, it used the GPT-3, a general-purpose model. The GPT line represents versatility, while the o lines are more inclined towards reasoning. The o3 was released in January 2025, while the o4-mini appeared three months later. These two models are also representatives of OpenAI in the first AI chess tournament in history.

Grok 4 is a creation by xAI, owned by the world's richest man Elon Musk. He said that Grok 4 had hardly learned chess before participating in this tournament.

The tournament was organized by Google on the Kaggle platform for three days between Aug. 5-7, with 8 AIs competing in a single-elimination format. The two Chinese representatives, Kimi K4 and Deepseek, were both eliminated early in the quarterfinals, with heavy losses. The remaining six representatives in the tournament are all from American companies. These are the strongest AIs in the world.

In the third-place match, Gemini 2.5 Pro beat o4-mini with a score of 3.5-0.5.

The LLMs participating in the tournament were all non-chess specialists. At the same time, another chess tournament was taking place between eight chess engines, which received less attention. These eight engines had Elo ratings ranging from 3,576 (Integral) to 3,731 (Stockfish). Some of them used AI algorithms to improve their performance. They were all far above human level.