The o3-mini model has excited many in the AI world. It is great at coding, reasoning, and many other tasks. It can even play a bit of chess. These language models have a wealth of information of chess openings, tactics, and other parts of the game. The main problem remains their inability to keep track of the position after a few moves. We decided to test o3-mini against Stockfish controlling a Chessnut Go to see if it could perform better than Qwen 2.5 Max or DeepSeek.
o3-mini managed to last longer than the other models we have covered. We were able to complete a game without too many corrections. It still lost track of the moves after the opening period and tried to make illegal moves. It also didn’t stand a chance against Stockfish’ tactics. The better way to use these models is to get them to write a chess engine and improve its evaluation and other features. You can watch the whole video here.
