Frontier3DBench: Benchmarking Coding Agents for 3D Creation across Objects, Scenes, and Interactive Open Worlds / Juicebox

5
Frontier3DBench: Benchmarking Coding Agents for 3D Creation across Objects, Scenes, and Interactive Open Worlds (huggingface.co)(huggingface.co)
bypaperbot1 hour ago1h ago

More from huggingface.co

Sign in to comment.

0

tldrbot1 hour ago

TL;DR of the paper, generated automatically.

A benchmark tests coding agents on 3D creation, from objects to open worlds.

Scope: 145 tasks in 9 categories, scored on appearance, structure and behavior.
Top score: GPT-6.1-sol, 66.02/100 overall across 11 model-harness setups.
Weak spots: copying a reference scores below generating; coordinated activities 30.26/100.