Gemini, a captivating but sometimes perplexing model, appears captivated by the examples it is given. Instead of applying general principles, it seems to mimic the provided examples, often overlooking the core instructions. Even the cutting-edge 2.5 models struggle with intricate tasks, falling short of the problem-solving abilities of GPT-4o-mini and o3-mini. Perhaps the model is focused on the surface, missing the deeper, more nuanced instructions.