WorldCoder-Bench

WorldCoder-Bench Benchmarking Physically Grounded 3D World Synthesis This repository contains the core task dataset for WorldCoder-Bench, a benchmark that evaluates how well large language models (LLMs) generate interactive, physically grounded 3D web scenes from natural language using Three.js. Generated programs must integrate 3D assets, obey spatial and physical constraints, and keep user-facing controls synchronized with hidden runtime state — mechanics that unfold inside… See the full description on the dataset page:

Source: Hugging Face Hub (shuolucs/WorldCoder-Bench). Metadata imported from the dataset’s Hub tags.