WorldTUI is a 360-million-parameter model that predicts how a terminal screen changes in response to keystrokes, without a real shell or filesystem. Its creator trained Qwen2.5-Coder on Docker sessions, added recovery from its own errors, parallel row generation and verification, and reduced the int4 demo to 208 MB. It runs locally in the browser through WebGPU and correctly handled 80 of 106 commands in the creator’s ShellBench. Unlike the prompt-based virtual terminals demonstrated in 2022, it learns screen updates themselves. However, it fails at tasks such as appending to files, nested paths and complex calculations, so it cannot replace a reliable execution environment.