OpenAI's o3 AI model defeated X's Grok 4 4-0 in the finals, winning the championship at the Kaggle Game Arena 2025 AI chess exhibition hosted by Google.
8 AI models participating in the Kaggle AI Chess Championship 2025. (Source: Chess.com)
Grok 4 was once considered the strongest candidate, but made many serious mistakes in the final match, such as losing pieces early and making unreasonable moves.
o3 showed excellent tactical analysis and understanding of chess strategy, especially in the middle and endgame. Grok 4, although strong in the early stages of the tournament, showed weakness in the final games and failed to show the ability to handle complex situations.
Elsewhere, Google's Gemini 2.5 Pro took third place after beating o4-mini 3.5-0.5. Despite the win, Gemini was judged to have played at a lower quality than o3, with many messy and error-prone games.
The Kaggle AI Challenge 2025 took place from August 5 to 7, bringing together eight large language models (LLMs) competing in a single-elimination format. The eight representatives included o3 and o4-mini from OpenAI, Grok 4, Gemini 2.5 Pro and Gemini 2.5 Flash from Google, Claude 4 Opus from Anthropic, and Deepseek R1 and Kimi R2 from China.
This is a Google experiment to evaluate the potential of AI in the fields of strategic thinking and general intelligence.
Source: https://vtcnews.vn/openai-o3-vo-dich-giai-co-vua-ai-kaggle-2025-ar958595.html
Comment (0)