# 3JSBench: A benchmark for LLM-generated Three.js objects

**URL:** <https://discourse.threejs.org/t/3jsbench-a-benchmark-for-llm-generated-three-js-objects/95137>\
**Category:** Showcase\
**Tags:** ai, geometry\
**Created:** [October 9, 2026, 5:07am UTC](https://discourse.threejs.org/t/3jsbench-a-benchmark-for-llm-generated-three-js-objects/95137 "2026-10-09T05:07:43Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![frvnk](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.threejs.org/frvnk/32/73276_2.png) [@frvnk](https://discourse.threejs.org/u/frvnk)\
**Post date:** [October 9, 2026, 5:07am UTC](https://discourse.threejs.org/t/3jsbench-a-benchmark-for-llm-generated-three-js-objects/95137/1 "2026-10-09T05:07:43Z")

</div>

Have you ever wondered which AI makes the best Three.js objects? Well I’d like to share [3jsbench.com](http://3jsbench.com)!

Having built [instaplay.ai](https://www.instaplay.ai/), an AI game generation platform home to over 40,000 games, we noticed a recurring problem: models can generate game mechanics for increasingly complex worlds, but 3D objects fail to meet the bar. We commonly see errors like flickering and invisible surfaces, as well as objects passing through each other.

You can find a leaderboard on which model’s 3D objects people prefer the most as well as which ones have the highest overall correctness. From the models we’ve tested we’ve found that most people prefer Opus 5.5’s generations the most while Astra passes in correctness most often.

x post: [link](https://x.com/generalcontext_/status/2108313483293147292?s=20)

contact: [frank@instaplay.ai](mailto:frank@instaplay.ai)

 ![image](https://canada1.discourse-cdn.com/flex035/uploads/threejs/original/3X/d/d/dda9ddcacb62b6f2469e4d233bd370cca8b16210.png)