OpenAI’s o3 model defeated xAI’s Grok 4 in a chatbot chess tournament, this contest help prove the advanced capabilities of everyday interactive agents which occurred nearly 30 years after a machine first beat a grandmaster.
The o3 model even beat Open AI’s o4 mini on its way to a final against Grok 4 on Google’s Kaggle Game Arena, which it won by four games to nought.
A face-off between Google’s Gemini 2.5 Pro and o4 mini for third place went the way of the search giant’s AI in a match where one game was drawn.
Kaggle explained a seeding system was used to ensure top tier chatbots did not meet before the final. Matches were streamed, with details of each model’s reasoning displayed and the rate of play optimised for viewing.
The results are significant because Kaggle’s Game Arena is a benchmarking platform using games to measure the performance of leading AI models, an approach it explained offers “a clear, unambiguous signal of success”.
Kaggle added games push AI models to demonstrate skills spanning “strategic reasoning, long-term planning and dynamic adaptation”.
Bloomberg noted the chess abilities of the AI models in the contest fall well short of the capabilities of IBM supercomputers which are now at a stage of teaching themselves complex games, however BBC News caveated this by highlighting the chatbots are everyday items not built from the ground up for a complex game.
Grandmaster battle
IBM’s Deep Blue became the first computer to win a chess game against a grandmaster in 1996.
The machine ultimately lost the contest against Garry Kasparov, but took the overall win in a rematch the following year.
Deep Blue used 32 processors and was capable of evaluating 200 million chess positions per second.
IBM states the machine was central to advancing the abilities of supercomputers to take on complex calculations needed for functions in pharmaceutical and financial sectors, along with analysing huge datasets and performing human gene research.
Source: Mobile World Live
Image Credit: OpenAI