Kaggle, in collaboration with Google Deepmindjust launched Game Arenaa new platform designed to understand better how artificial intelligences think. To inaugurate the project, from5 to 7 August 2025 a hold of a chess tournament between ai which features some of the Most advanced linguistic models in the worldincluding chatgpt, Gemini, Claude and others.
It is a challenge a Direct eliminationwhere the Ai face one against one. Unlike other systems created only to play, these models are generalistsdesigned to write texts, program, solve problems. And now they are trying to chess.
The participants in the tournament
To compete on the chessboard there will be eight models of AI developed by some of the most powerful realities in the field of artificial intelligence:
To tell the tournament to the public there will be well -known faces in the world of chess online:
There ranking will be updated in real time On Kaggle, in Elo style, so it will be easy to understand which AI is truly the strongest in the game of chess.
These are not born to play. I’m not like Alphazerthe Google Deepmind system that made history in 2017 by beating the Stockfish engine, one of the strongest ever. Alphazero learned alone to play, simply by playing millions of games against himself. But it has never been made public.
The models that we will see on the field in the Kagger tournament, on the other hand, are ai generalist: they know how to write, answer questions, generate code, but often make chess errors. Some of them make illegal moves, or surrender without reason.
The interesting thing, however, is that these are able to explain why they make a certain move. According to Google, this feature helps to really understand how their way of thinking works, going beyond the simple score of a game. Kaggle focuses on this: to make visible the evolution of the strategic thought of the AI.
Chess today, complex simulations tomorrow
Game Arena will not stop at board games. Kagger, owned by Google, plans to extend the platform to simulated environments, multiplayer games and realistic scenarios. Everything will be open, accessible and designed to help developers and researchers to understand how they behave and adapt to the in different contexts.
For now, however, the spotlights are focused on chess. And even if none of these models still reaches the level of an expert human master, the challenge is fascinating. Because it allows us to observe closely how a car makes decisions, it is wrong, learns and adapts. And this applies more than a victory over the chessboard.
