LLMs don’t currently “think”. So I assume a slow response is just a system load issue.