Which LLM does understand chess best?

[BeyondCritics]

I am interested in using LLMs to work with chess related content, e.g. chess history, important games, basic chess strategy. As of January 2026, is there a model you could suggest? Does it even work?

[FireDragon761138]

I am interested in using LLMs to work with chess related content, e.g. chess history, important games, basic chess strategy. As of January 2026, is there a model you could suggest? Does it even work?

I've used ChatGPT and Claude, particularly with Retrieval Augmented Generation (Projects, CustomGPT's based on classic texts like Silman or Stean's books). DeepSeek is also surprisingly good, and it's completely free for the moment.

It's educational and amusing to just paste in a game an engine has analyzed. I recommend using a more positional chess engine for the initial analysis (Dragon, Komodo, or Lc0, etc.), it will help cue in the language model towards positional features- Stockfish isn't so good as it tends to tilt the model towards tactics that don't necessarily parse so well into strategic concepts. It's been a big help in transitioning into a more positional style. The LLM's can also be instructed to balance its own judgement with the engine output, which is helpful in identifying play that is practically good rather than merely engine-approved.

[towforce]

I am interested in using LLMs to work with chess related content, e.g. chess history, important games, basic chess strategy. As of January 2026, is there a model you could suggest? Does it even work?

General advice: get the best chatbot you can afford. However, different chatbots very clearly have different strengths, so you'll need to experiment to find out which one is best at this specific task. I'm guessing you were hoping that somebody else had already done that.