Get the latest tech news

SketchAgentBench – evaluate LLMs on 3D spatial reasoning


One photo. One prompt. Code, render, inspect, refine. 11 models, 25 photos, 275 runs—each scored on its final 3D scene.

None

Get the Android app

Or read this on Hacker News

Read more on:

Photo of LLMs

LLMs

Photo of reasoning

reasoning

Photo of 3D spatial

3D spatial

Related news:

News photo

Memorilla: Latent Semantic Memory for LLMs

News photo

When Modernization Breaks Working Code: A Study of LLMs on Legacy PHP

News photo

So I guess everyone uses LLMs to think now