Humor Arena Benchmark is a tool that measures the funniness of jokes generated by AI models. It uses a dataset of human ratings to detect which jokes people find funniest, acknowledging that humor has both objective and subjective components. This tool is for creators, marketers, and anyone interested in using AI for humorous content. It helps you understand which AI models are better at generating genuinely funny jokes, allowing you to select the best tools for your creative projects. The benchmark also notes that longer thinking times can slightly improve joke quality and that absurdity often correlates negatively with joke quality. While the tool can detect humor, it acknowledges that subjective humor is outside its current scope. The models tested never refused to generate jokes, even with dark prompts, and the benchmark provides insights into the performance of various AI models in humor generation.