On August 6, 2025, the first AI Chess Match hosted by Google DeepMind and Kaggle entered the semifinal stage on the Kaggle Game Arena platform. This competition brought together top AI models from around the world, competing in a single-elimination format through pure text interaction, aiming to test the capabilities of large models in complex decision-making and logical reasoning.
In the two highlight matches of the semifinals, xAI's Grok4 unexpectedly defeated Google's Gemini 2.5 Pro, while OpenAI's O3 won against its sibling model O4-mini in an "internal battle," drawing widespread attention.
Grok4 Unexpectedly Defeats Gemini 2.5 Pro
In the first match of the semifinals, xAI's Grok4 faced Google's flagship model, Gemini 2.5 Pro. This match was considered a showdown of technological approaches, with Gemini 2.5 Pro previously regarded as a favorite to win due to its strong tactical vision and stable performance. However, Grok4 won decisively with a score of 4:0, showcasing remarkable strategic analysis abilities.
During the game, Grok4 demonstrated steady play in the opening phase, accurately grasping the board situation and forcing Gemini 2.5 Pro into a passive position multiple times through clever tactical arrangements. Although Gemini 2.5 Pro had previously defeated Anthropic's Claude Opus 4 with a 4:0 score in the first round, showing excellent chess skills, it appeared overly lengthy in its analysis during this match, failing to respond effectively to the opponent's attack rhythm. Grok4, on the other hand, secured victory with efficient reasoning and deep insights into the game state, making critical tactical breakthroughs at key moments.
After the match, online reactions to Grok4's performance were enthusiastic. Analysts believe that Grok4's fast adaptability and precise judgment in dynamic games may have been the key to its victory. This win has made Grok4 a strong favorite for the final, drawing much anticipation.
O3 Overcomes O4-mini, OpenAI's Internal Battle Ends in Victory
In another semifinal match, OpenAI's O3 faced off against O4-mini in a highly anticipated "internal battle." O3 won with a score of 4:0, advancing to the final. The game lasted a shorter time, with a relatively smooth progression, as O3 displayed stronger stability and control over the game state.
O4-mini had previously defeated DeepSeek R1 with a 4:0 score in the first round, demonstrating solid skills, especially with two instances of "checkmate" that left a strong impression. However, during its match against O3, although O4-mini showed some highlights in the opening, its judgment errors in the middle game caused it to gradually lose the initiative. O3, on the other hand, secured victory with more efficient reasoning and precise control over the game state, playing steadily and firmly.
Notably, neither side made serious mistakes during the match, with high-quality gameplay reflecting OpenAI's deep expertise in chess strategy and reasoning. O3's victory further solidified its competitiveness in the field of AI chess.
Competition Background and Outlook
This AI Chess Match aims to test the logical reasoning and strategic capabilities of large models through dynamic games. Participating models include Google's Gemini series, OpenAI's O3 and O4-mini, Anthropic's Claude Opus 4, DeepSeek's R1, and Kimi's K2. Throughout the competition, external tools such as the Stockfish engine are prohibited, relying solely on the real-time deduction ability of AI models. The results of the semifinals show that Grok4 and O3 will face off in the final, where the champion of the first AI Chess Match will be determined.
As the competition enters its most intense stage, Grok4's strong rise and O3's stable performance add more suspense to the final. Will xAI's "rebellious" AI come out on top, or will OpenAI's top model defend its glory? The answer will be revealed on August 7.