Skip to content

Dataset · HelenPeng

Executable 3D world code generation benchmark

HelenPeng/ExeWorldBench

An English multimodal benchmark for evaluating whether models can write working 3D scene code.

What's inside

The dataset contains between 1,000 and 10,000 English text and image examples. Unlike standard visual generation, the benchmark tests models' ability to understand world structure and recreate a working 3D environment as executable code.

What you can do with it

  • Evaluating multimodal models (VLMs) on text-and-image 3D scene generation tasks
  • Comparing LLMs on their ability to write correct executable code for 3D environments
  • Testing AI spatial reasoning and understanding of physics/object structure
Tasks
Text generation, Image text → text
Languages
English
Size
1K–10K rows
Downloads this month
1.6K
Likes
41

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("HelenPeng/ExeWorldBench")