They are scraped from the web, and discussed on Reddit. So, they are definitely in the training data. Despite that, the non-reasoning LLMs struggle to solve the

They are scraped from the web, and discussed on Reddit. So, they are definitely in the training data. Despite that, the non-reasoning LLMs struggle to solve them. There are however new problems each week, and released every week. So, we can safely assume the latest problems are decontaminated. It remains to be seen if and how performance drops on the problems released in 2025. (Not enough problems yet to tell.)