perhaps they forgot to set thinking mode in the LLM payload