I test Qwen models regularly. They are very good for English and I'm guessing Chinese, but much worse for non-English (specifically, Polish).