OpenAI Defeats Musk’s Grok in First AI Chess Tournament Final

The first-ever three-day Kaggle AI Exhibition Tournament concluded in Beijing, with OpenAI o3 convincingly defeating Elon Musk’s xAI Grok 4 4-0, achieving an average accuracy of over 91% over 12 games.

The competition took place on August 5–7, 2025, on Google’s Kaggle platform. In the final, Grok made a series of blunders, including losing her queen repeatedly, while OpenAI o3 acted with maximum precision and ruthlessness. Grok looked like the favorite going into the semifinals, but the final day was dominated by o3.

Google Gemini took third place, defeating o4-mini 3.5–0.5. In total, the tournament featured eight major language models from Anthropic, Google, OpenAI, xAI, and Chinese developers DeepSeek and Moonshot AI (which were eliminated in the quarterfinals).

Developers use such tournaments to test their models’ abilities in tasks that require strategic thinking and working with clear rules, such as chess or programming. Elon Musk used to co-found OpenAI, but left the organization to start his own company, xAI. Before the finals, he noted on X (Twitter) that his team “almost did not spend any effort on chess.”

OpenAI o3’s victory in the chess tournament revealed a significant gap in the strategic abilities of language models and highlighted the role of specialization in AI training. The competition proved that even in tightly structured games, the best results are achieved by models that combine accuracy with adaptability.