Overview

• The three-day exhibition on Google’s Kaggle Game Arena concluded on August 7 with OpenAI’s o3 emerging unbeaten after a 4–0 final sweep of xAI’s Grok 4.
• Eight general-purpose LLMs from OpenAI, xAI, Google, Anthropic, DeepSeek and Moonshot AI competed without any specialized chess training.
• Observers noted the models performed at roughly 800–1200 ELO strength, oscillating between solid moves and inexplicable blunders such as repeated queen losses.
• World champion Magnus Carlsen provided live commentary, quipping that the matches resembled “kids’ games” and sharing personal anecdotes about playing with tech leaders.
• The tournament highlighted both the potential and the limitations of general-purpose AI in structured, turn-based tasks, as consistency gaps persisted.