BlenderWorld benchmarks frontier models on real-world 3D creation
Original title: BlenderWorld: Benchmarking Visual Agents for 3D Creation from Real-World References
GPT-6-Astra scored 79.9 overall, ahead of Opus-5 at 73.0 and GPT-5.6-Sol at 67.7.
BlenderWorld is a benchmark of 223 realistic 3D creation tasks across architecture, products, natural environments, and scientific visualization. Inputs include text, image, video, and structured specifications. Submitted Blender projects are rendered from multiple viewpoints and scored against 2,651 binary rubrics plus visual quality scores. Among thirteen frontier models, GPT-6-Astra led with a fused score of 79.9. The authors say models still struggle with fine-grained 3D spatial coherence and structural reproduction.
- The benchmark has 223 tasks and 2,651 task-specific binary rubrics.
- Thirteen frontier models were evaluated.
- GPT-6-Astra scored 79.9, Opus-5 73.0, and GPT-5.6-Sol 67.7.
- Inputs cover text, image, video, and structured specifications.
- Authors report weak fine-grained spatial coherence in Blender.
- Products
- BlenderWorldBlender
- AI models
- GPT-6-AstraOpus-5GPT-5.6-Sol
From the original article
1 min read at huggingface.co
Recent advances in multimodal large language models (MLLMs) have enabled them to act as autonomous agents, creating and editing 3D models via code or direct software manipulation. In practical workflows, these models must infer object shapes, materials, spatial layouts, and lighting from diverse references while strictly following instructions to reconstruct, modify, or design 3D content.
By Xinyu Fang, Yuchi Wang, Weihao Yu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.