According to God of Prompt on Twitter, leading frontier AI models such as GPT4, Claude, and Gemini demonstrate minimal overfitting when tested on contamination-free datasets, indicating genuine reasoning capabilities. However, as reported by God of Prompt, many mid-tier open-source models exhibit widespread contamination issues across various sizes and versions. This suggests that while top-tier proprietary models maintain high data integrity and robust reasoning skills, open-source alternatives may face significant challenges in ensuring clean training data and preventing overfitting, which could impact their reliability and business adoption.