Kaggle has launched Community Benchmarks, allowing AI developers to create and share custom benchmarks for evaluating their models. This move builds on last year's introduction of Kaggle Benchmarks, which provided access to evaluations from notable research groups like Meta and Google. The rapid evolution of AI capabilities has made traditional evaluation methods, such as relying solely on accuracy scores from static datasets, inadequate. As large language models (LLMs) become more complex, the need for a more dynamic evaluation framework is clear. Community Benchmarks enable users to design specific tasks that assess various AI functions, such as multi-step reasoning, code generation, and image recognition. Once these tasks are established, they can be grouped into a benchmark to evaluate multiple models and generate a performance leaderboard. Benefits include free access to top models from organizations like Google and Anthropic, reproducibility of results, and the ability to test complex interactions involving multi-modal inputs and tool usage. The introduction of the kaggle-benchmarks SDK supports users in creating and refining their benchmarks efficiently. Resources like the Benchmarks Cookbook and example tasks provide guidance for users to jumpstart their projects. The platform empowers developers not only to test AI models but also to influence the future of AI evaluation, pushing the boundaries of what these systems can achieve.