Has anyone else also noticed that AI LLMs get lazier the more complex/longer the task is?

I have a prompt to analyze and compare different news paper articles and the first article has 4 very long paragraphs, and the last one only has a few short bullet points. This is Grok 4, but this behavior is the same with every LLM model I've tested.

I haven't found a solution to this problem, only a workaround by prompting each article individually. However, this lacks the context of other articles, which is needed for a comparison.