Atlas

Benchmarks

← All benchmarks

BlenderWorld Text

Code · 2026-10-14

The 78 BlenderWorld 3D creation tasks specified in text alone. A coding agent works in an isolated GPU container with shell, file and image tools plus a Blender MCP server, and must deliver an editable .blend project with the requested renders or video. A GPT-5.6 Terra judge scores each task as the equal-weight mean of a weighted binary-rubric pass rate and eight 1-10 visual-dimension ratings, both on 0-100; a missing .blend scores zero. The score is the mean fused task score over these tasks.

Top models (higher is better)

ModelScore
GPT-6 Astra78.6
Claude Opus 574.8
GPT-5.6 Sol71.7
DeepSeek V4.1 Flash66.0
GLM-5.3 Flash63.1
Grok 4.660.0
Opus 4.858.6
Qwen3.8 27B46.1
Kimi K341.3
Gemini 3.8 Flash35.9
Loading Atlas data…