Artificial intelligence has made remarkable strides in solving complex mathematical problems. But can AI be truly *creative* in math, discovering novel solutions and pushing the boundaries of mathematical knowledge? A new study explores this question by assessing the creativity of Large Language Models (LLMs) in proposing innovative solutions to mathematical challenges. Researchers introduced a novel framework and benchmark called CREATIVEMATH, containing problems ranging from middle school level to the complexities of Olympiad competitions. The goal? To see if LLMs could devise new solutions after being shown existing ones. The results reveal a fascinating dynamic. While LLMs generally excel at standard math problems, their creative abilities vary significantly. Notably, Gemini 1.5 Pro shone in generating unique solutions, often deviating from the provided examples. Interestingly, providing more reference solutions initially boosted accuracy, with Gemini achieving perfect scores when given four prior solutions. However, this abundance of examples also seemed to stifle creativity, suggesting a trade-off between leveraging existing knowledge and fostering original thought. As the mathematical problems became harder, LLMs struggled with accuracy, but their successful attempts were more likely to be innovative. This implies that tougher challenges can spur AI creativity. Furthermore, by analyzing the similarities between solutions from different LLMs, researchers found that some models, like Llama 3 and Yi, explored diverse approaches, while others, such as Mixtral, tended to produce similar solutions. This highlights the value of using a variety of LLMs to maximize the potential for innovative solutions. This research offers a glimpse into the evolving landscape of AI and its potential to reshape how we approach mathematical discovery. While challenges remain, the potential for AI to assist in, and even drive, mathematical innovation is becoming increasingly clear.