Can large language models (like ChatGPT and DeepSeek) play Xiangqi?
As the name suggests, a large language model (LLM) is mainly built to process language. It cannot play Xiangqi, just as some people who lack expert knowledge can still talk confidently about wrong things. (Unless it calls a separate Xiangqi engine, just as a human who can't play Xiangqi can still use a Xiangqi engine.)
Each field has its specialists. If you want to play Xiangqi, use a dedicated Xiangqi engine AI.
Unlike dedicated Xiangqi engines, LLMs have not been given the rules of the game, and LLMs themselves hallucinate, so they do not strictly follow the rules and very easily play illegal moves and move pieces at random. That said, GPT may have a dedicated move check for chess, so hallucinations there are rare.