Comment by bArray

It could be that the prompt is accidentally (or purposefully) more optimised for Kimi K2, or that Kimi K2 is better trained on this particular data. LLM's need "prompt engineers" for a reason to get the most out of a particular model.