A unique and powerful approach to understanding AI capabilities, Kaggle Game Arena has teamed up with Google DeepMind to unveil a new platform where major AI models compete directly in chess and other games.
To get things started, Kaggle will hold a three-day AI exhibition chest tournament on August 5th-7th, featuring eight world-leading AI models. Competitors include Google, Openai, the Large-scale Language Models of Humanity (LLM) and other labs.
- Gemini 2.5 Pro (Google)
- Gemini 2.5 Flash (Google)
- O3 (Openai)
- o4-mini (openai)
- Claude 4 Opus (Anthropology)
- Grok 4 (Xai)
- Deepseek R1
- Kimi K2 (MoonshotAI)
Introducing Kaggle Game Arena: A new open benchmark platform where top AI models compete in complex and strategic games with streamed matchups. We chart new frontiers for reliable AI ratings, and it starts with chess. This is the classic proof foundation for system intelligence. pic.twitter.com/ohbwbnnqtn
– kaggle (@kaggle)) August 4, 2025
Chess.com helps Kaggle bring events to a larger audience with event page coverage and daily news reports. The launch will also be in partnership with three biggest names: Take Take and Chess.
- GM hikaru nakamura provides live, daily commentary on his Twitch stream
- Im Levy Rozman, also known as Gothamchess, provides daily summaries and analysis on his YouTube channel.
- The tournament concludes with a summary from GM Magnus Carlsen on the Take Take website and shares insights on AI vs. AI action.
The tournament follows a single removal format, using seeds based on preliminary test matches. Following the tournament, a continuous updated leaderboard with rankings like ELO will be available in Kaggle to track the performance of all participating models. This gives the general public a clear way to see which AI is the best in chess.
Kaggle is partnering with Google Deepmind, which revolutionized chess in 2017 with the release of its chess play computer program Alphazero. This was when Alphazero, which was not made public, reached a level that was almost unthinkable when he reached a level that was almost unthinkable by playing millions of games using only reinforcement learning and using “millions of games” for four hours.
Alphazero is famous for defeating Stockfish, the world's strongest chessplay computer program, in a one-sided 100 game match. He will win his second match against Stockfish in late 2018 with over 1,000 games.
However, the AI models competing in Kaggle are not close to the level of Alphazero. Unlike Alphazero, these LLMs are “generic” tools for coding, writing and reasoning about the world, and are not specifically programmed to play chess. As Chess.com details, these models are still learning and are known to carry out both illegal moves and absurd resignations.
The launch of the gaming platform offers a unique opportunity to observe the evolution and improvements of various AI models. It is estimated that many LLMs, such as Openai's ChatGpt and Google's Gemini, are currently playing at the amateur player level.
Although they lack the strength of play, these models can provide “inference” behind each movement. According to Google, this allows you to “beyond your static score and see how AI works in a dynamic and competitive environment.” Kaggle hopes that the initiative will provide insight into how AI is “thinking” the model and the future of AI's strategic intelligence.
Kaggle Game Arena is expanding beyond board games as billion-dollar companies plan to incorporate more complex, multiplayer, real-world simulation environments. Kaggle will open throughout the AI community, allowing you to have a transparent understanding of model performance across a variety of games.
