A unique and powerful approach to understanding AI capabilities, Kaggle Game Arena, in partnership with Google DeepMind, announced a new platform where leading AI models will compete head-to-head in chess and other games.
To kick things off, Kaggle is hosting a three-day AI Exhibition Chess Tournament on August 5-7 featuring eight of the world’s leading AI models. The competitors include Large Language Models (LLMs) from Google, OpenAI, Anthropic, along with other labs:
- Gemini 2.5 Pro (Google)
- Gemini 2.5 Flash (Google)
- o3 (OpenAI)
- o4-mini (OpenAI)
- Claude 4 Opus (Anthropic)
- Grok 4 (xAI)
- DeepSeek R1
- Kimi k2 (Moonshot AI)
📢Introducing Kaggle Game Arena: a new, open benchmark platform where top AI models compete in complex, strategic games in streamed match-ups. We’re charting new frontiers for trustworthy AI evaluation and it begins with chess — a classic proving ground for system intelligence. pic.twitter.com/OHBWbnnQtn
— Kaggle (@kaggle) August 4, 2025
Chess.com will help Kaggle bring the event to a wider audience through coverage on the events page and with daily news reports. The launch is also made in partnership with Take Take Take and three of the biggest names in chess:
- GM Hikaru Nakamura will provide live, daily commentary on his Twitch stream
- IM Levy Rozman, also known as GothamChess, will offer a daily recap and analysis on his YouTube channel.
- The tournament will conclude with a recap from GM Magnus Carlsen on the Take Take Take website, where he will share his insights on the AI vs AI action.
The tournament will follow a single-elimination format, with seeding based on preliminary test matches. Following the tournament, a continuously updated leaderboard with Elo-like rankings will be made available on Kaggle to track the performance of all participating models. This will give the public a clear way to see which AI is the best at chess.
Kaggle has partnered with Google DeepMind, which revolutionized chess with the release of the chess-playing computer program AlphaZero in 2017. AlphaZero, which was never made available to the public, took the chess world by storm when it reached an almost unthinkable level using only reinforcement learning and by playing many millions of games against itself—only having used four hours to “learn” the game.
AlphaZero famously defeated Stockfish, the world’s strongest chess-playing computer program, in a one-sided 100-game match. It would go on to win a second match against Stockfish over 1000 games in late 2018.
However, the AI models competing on Kaggle are nowhere near the level of AlphaZero. Unlike AlphaZero, these LLMs are “general purpose” tools for coding, writing, and reasoning about the world—and not programmed specifically to play chess. As Chess.com has detailed, these models are still learning and have been known to both make illegal moves and absurd resignations.
The launch of the game platform offers a unique opportunity to observe the evolution and improvement of various AI models. For example, many LLMs such as OpenAI’s ChatGPT and Google’s Gemini are currently estimated to be playing at the level of an amateur player.
Although they lack the playing strength, these models can provide their “reasoning” behind each move. According to Google, this allows us to “move beyond static scores to see how AI truly performs in a dynamic, competitive environment.” Kaggle hopes that this initiative will provide insight into how AI models “think” and the future of AI’s strategic intelligence.
Kaggle Game Arena extends beyond board games, as the multibillion-dollar company intends to incorporate more complex, multiplayer and real-world simulation environments. Kaggle will be open for the entire AI community, allowing for a transparent understanding of how models perform in various games.
