Reportedly: Alibaba Qwen open-sources BlenderWorld, a 223-task benchmark for 3D creation agents in Blender
Reported by IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Alibaba Qwen released BlenderWorld on October 14, an MIT-licensed benchmark with 223 tasks and 2,651 criteria for evaluating visual agents on real 3D creation in Blender. Code and data are on GitHub, Hugging Face, and ModelScope, IT Home reported.
China context
- Original name
- 阿里巴巴
- Outside China
- Open weights · ithome.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can use the MIT-licensed benchmark to test their own 3D agents against Qwen's standard, with code and data available on GitHub and Hugging Face.
- For investors
- The open-source release signals Alibaba's commitment to developer ecosystems and may drive adoption of Qwen models in 3D content creation tools.
Translated from Chinese. Quotes and facts link to the original sources.
BlenderWorld evaluates AI agents on producing editable .blend files with materials, geometry, lighting, and cameras, not just images. It covers 9 categories and 22 subcategories, with tasks given as text, image+text, video+text, or spec files. Scoring combines rule compliance and visual quality, each weighted 50%.
The benchmark requires agents to output editable Blender project files and renders from 8 adaptive viewpoints, with 2,651 binary scoring criteria and 8 visual quality metrics. This tests multi-modal understanding and tool use in a 3D environment.
3D designers and studios gain a standardized way to compare AI tools for Blender workflows, potentially reducing evaluation costs. Alibaba's open-source release pressures other AI labs to match or exceed performance on this benchmark.
For developers, BlenderWorld provides a free, open-source evaluation suite to benchmark and improve 3D creation agents. For Alibaba, it positions Qwen as a leader in embodied AI tooling and may attract enterprise partnerships.
Watch for model submissions and leaderboard updates on the BlenderWorld repository, and whether other labs adopt or extend the benchmark.