The 3D Arena leaderboard evaluates leading generative 3D models. It is inspired by TTS-Arena and Chatbot Arena.

Read more about the platform and methodology in the official paper: 3D Arena: An Open Platform for Generative 3D Evaluation.

Motivation

Evaluating generated 3D quality quantitatively is challenging, and there is no standard practice for reporting critical factors like mesh topology in real-world applications.

This leaderboard displays usable, final results side-by-side for the community to rank.

Criteria

Models must handle image (.png, .jpg) inputs and produce mesh (.obj, .glb) or splat (.splat, .ply) outputs. They should run end-to-end without human intervention, including UV unwrapping, texture mapping, and other post-processing.

To preview how your results will look in the leaderboard, upload them to any gradio/model3d viewer.

Contributing

All inputs and outputs are publicly available here. Input image URLs are provided here.

To contribute a model, run your model on all input images and create a pull request with your model's output. Refer to other examples in the dataset to understand the expected format.

Citing

            
                @misc{ebert20253darenaopenplatform,
                    title = {3D Arena: An Open Platform for Generative 3D Evaluation}, 
                    author = {Dylan Ebert},
                    year = {2025},
                    eprint = {2506.18787},
                    archivePrefix = {arXiv},
                    primaryClass = {cs.CV},
                    url = {https://arxiv.org/abs/2506.18787}, 
                }