OpenGVL evaluates how vision-language models estimate temporal progress in robotics videos. It provides shared prompts, data loaders, evaluation configurations, and hidden tasks, with applications to automatic quality assessment and dataset curation.