3DCodeVerse: Exploring 3D Worlds as Code — Ecosystem for Editable and Verifiable 3D Content via Coding Agents

Forum topic · 小凯 · 2026-10-17

Summary

3DCodeVerse (arXiv 2610.22341) is a research ecosystem that enables coding agents to create editable and verifiable 3D content at scale. It consists of three components: 3DCodeVerse-2M, a large-scale curated dataset of standalone executable 3D code built from open-source data, converted specialized representations, and agent-generated programs, each packaged with source files, assets, descriptions, and verification records; the 3DCodeVerse Harness, a multi-agent directed acyclic graph framework that automatically evaluates and revises generated programs using execution checks, spatial measurements, and visual feedback without human input; and practical guidance for 3D coding. Experiments show that fine-tuning open-source models on the dataset significantly improves their 3D coding ability, and the framework achieves a 90% win rate in pairwise human evaluation against Claude Design (3D Object). Fields: computer vision; authors include Yipeng Gao, Jiebo Luo, Laurent Itti, and others.

3DCodeVerse: Exploring 3D Worlds as Code

Research area: Computer Vision (CV)

arXiv: 2610.22341

Release date: 2026-10-15

Authors: Yipeng Gao, Ziyao Zeng, Jinfa Huang, Yijiang Li, Haolin Xiong, Wang Qin, Xinlei Yu, Youheng Yao, Jie Zhang, Kun-Yu Lin, Rong Liu, Yinyuan Zhao, Yajie Zhao, Meiqi Guo, Jiebo Luo, Lei Shu, Yunhao Ge, Xinyu Hu, Laurent Itti

Overview

We introduce 3DCodeVerse, a research ecosystem that enables coding agents to create editable and verifiable 3D content at scale. 3DCodeVerse comprises large-scale, high-quality datasets of executable 3D code, a coding harness for spatial reasoning and automated verification, and practical guidance for 3D coding.

3DCodeVerse-2M

First, the authors construct 3DCodeVerse-2M, a large-scale curated dataset of standalone 3D executable code. The corpus is built by:

Collecting open-source data
Converting specialized representations
Generating new programs with coding agents

Each program is packaged with the source files and associated assets required for execution in its declared environment, along with available descriptions and verification records.

3DCodeVerse Harness

Second, the authors develop the 3DCodeVerse Harness, a multi-agent directed acyclic graph (DAG) framework that uses execution checks, spatial measurements, and visual feedback to automatically evaluate and revise generated programs — without requiring human feedback.

Results

Extensive experiments show that fine-tuning open-source models on the 3DCodeVerse-2M dataset significantly improves their 3D coding capabilities.
The framework achieves a 90% win rate in pairwise human evaluation against Claude Design (3D Object).

---

*Auto-collected on 2026-10-17.*