Seems there are issues with reasoning models like **qwen3**. The program don't manage the tags \<think\> and \</think\>, so the output is very messy.