Alibaba Qwen Open-Sources BlenderWorld to Test AI 3D Creation Skills

Alibaba's Qwen team has open-sourced BlenderWorld, a benchmark that tests how well AI agents handle real 3D creation tasks in Blender. The code and dataset are now available on GitHub, Hugging Face, and ModelScope.

Unlike just generating a picture, this benchmark asks AI to deliver an editable .blend project file and build 3D scenes with materials, geometry, lighting, and cameras that designers can keep editing. It includes 223 tasks and 2,651 criteria across 9 categories, such as products, buildings, characters, and animation.

The scoring is strict: the system renders submitted projects from 8 adaptive angles, runs scene checks, and applies 2,651 task-specific binary criteria plus 8 visual quality scores. Rule compliance and visual quality each count for 50%. The project supports any generation framework and needs Linux and Python 3.10 or above.