AI‑Scientists have emerged, yet they lack a world in which to grow. Human researchers benefit from centuries of open problems and shared ideas recorded in the literature—readable but not directly runnable. To address this, we built Scientist World, turning every research paper into an executable environment. Our autonomous pipeline extracts each paper's open‑ended question and reproduces the proposed method as a baseline; an AI‑Scientist then enters the environment and is evaluated against the original authors' results. We release the first 1,000 verified environments. Using this world as an evaluation benchmark, frontier AI‑Scientists already surpass human researchers in 74% of cases. The record contains 6,400 hours of research traces, documenting every idea, experiment, failure, and success. As a training ground, we can boost a small model’s research ideation capability by 158%, bringing it on par with frontier models. At last, a dedicated world for AI‑Scientists has arrived and will continue to grow alongside science itself.\n\nReview: Scientist World transforms static papers into interactive experiments, offering a unified, reproducible benchmark for assessing and advancing AI‑Scientists, and highlights the emerging superiority of AI in scientific discovery.