Grok 4 has generated a lot of excitement among AI enthusiasts. It is topping the charts in various categories. These LLMs are not designed to play chess. In fact, most of them struggle to make valid moves after 13 or 14 moves based on our previous tests. We decided to see how smart Grok 4 can be in this area. We put it against Chessnut EVO’s Stockfish 17 at max level. We found Grok 4 to be much slower than every thinking model we have tested when playing the game. It checks Lichess and other sites to ensure accurate moves. Unlike Gemini, o3, and earlier versions of Grok, Grok 4 didn’t make any invalid moves.
As you can see in the above video, Grok needed a lot of time to make just a bunch of moves. It played much better than other models but still got crushed by Stockfish 17. Since this AI can use tools and do searches, it can serve as a great educational tool to explain ideas behind openings and certain moves. For example, we can see these smart chessboards adopting Grok or future versions of GPT to offer more personalized feedback to players.
