Fuzzing is a critical technique in software testing, aiming to uncover vulnerabilities by generating diverse inputs and observing program behaviors. During fuzzing, the seed corpus expands as new inputs that cover previously unseen edges in the program’s execution path are added. However, over time, many seeds in the corpus become redundant, as they can be effectively replaced by better-performing seeds. Retaining these outdated seeds slows the fuzzing process, as computational resources are still spent on them. To address this issue, we propose and implement a dynamic corpus minimization strategy, where the fuzzer periodically prunes the corpus to remove seeds that are unlikely to contribute to further code coverage. By focusing on a refined set of high-value seeds, our approach accelerates fuzzing and improves code coverage. We performed our experimental evaluation during the SBFT’25 Fuzzing Competition.