Imagine a world where AIs battle each other to determine who's the smartest. That's the innovative idea behind Auto-Arena, a new framework designed to automatically evaluate Large Language Models (LLMs). Forget static tests and expensive human evaluations – Auto-Arena pits LLMs against each other in multi-round debates, pushing them to their limits and revealing their true strengths and weaknesses. Like a digital debate club, two LLM candidates answer questions, critique each other's responses, and even devise follow-up questions to expose flaws in their opponent's reasoning. A committee of LLM judges then deliberates, mimicking human voting to decide the winner. This process not only automates evaluation but also reveals fascinating insights into LLM behavior. Researchers have observed LLMs demonstrating competitive strategies, learning from their opponents, and even exhibiting self-improvement during these battles. In tests with 15 leading LLMs, Auto-Arena's results closely mirrored human preferences, achieving a remarkable 92.14% correlation. This automated approach offers a promising alternative to traditional methods, providing faster, more efficient, and potentially more insightful LLM evaluations. The future of AI assessment may just lie in the arena of automated debate.