At the inaugural "AI Chess Exhibition" held on Google's Kaggle platform, OpenAI's o3 model overwhelmingly won the championship, defeating xAI's Grok 4 4-0 in the final, securing a complete victory in all three matches. The three-day event attracted eight AI models, including Google Gemini and Anthropic Claude, and required participating models to rely solely on publicly available chess knowledge acquired beforehand, without any specific training.
Although Grok 4 demonstrated strong performance in the semifinals, narrowly defeating Gemini 2.5 Pro 3-2, its tactical flaws were repeatedly exploited in the final by o3, who accurately exploited and repeatedly defeated it. Chess Grandmaster Hikaru Nakamura commented on the match that Grok 4's mistakes were "obvious like a beginner's," while o3 demonstrated stable computational prowess, despite both players possessing only amateur-level skills of 800 Elo. Notably, o3 also swept its fellow team, o4 mini, 4-0 in the semifinals, never losing a single game, ultimately securing the championship with a perfect 12-0 record.
Compared to AI specifically designed for chess (such as Stockfish), general-purpose models still lack the ability to handle endgames. Carlsen, the world's top chess player, likened them to "good at collecting pieces but not at checkmate." This tournament not only revealed the potential of general-purpose AI for complex reasoning tasks but also provided a new battlefield for technological competition among tech giants. The battle between OpenAI and xAI has been seen as a proxy war between Ultraman and Elon Musk.
