---
format: "aidr-story-markdown/v1"
id: "2fdc29bf5084a19a3c467c9a4fa444c580a0039656744f9e51f5ba03bb509120"
canonical_url: "https://aidr.today/2fdc29bf?lang=en"
title: "SketchAgentBench – evaluate LLMs on 3D spatial reasoning"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T20:28:24.000Z"
category: "Models"
topics: ["reasoning"]
source_urls: ["https://sketchagent.dev/","https://news.ycombinator.com/item?id=50398010"]
summary: "SketchAgentBench evaluates large language models on 3D spatial reasoning using one photo and one prompt, with 11 models tested across 25 photos and 275 runs, each scored on its final 3D scene."
---

# SketchAgentBench – evaluate LLMs on 3D spatial reasoning

> [Open the canonical story](<https://aidr.today/2fdc29bf?lang=en>)

**Published:** 2026-11-06T20:28:24.000Z
**Category:** Models
**Topics:** reasoning

## Summary

SketchAgentBench evaluates large language models on 3D spatial reasoning using one photo and one prompt, with 11 models tested across 25 photos and 275 runs, each scored on its final 3D scene\.

## Sources

- [Story source](<https://sketchagent.dev/>)
- [Discussion](<https://news.ycombinator.com/item?id=50398010>)