VoxelCodeBench: leaderboard

VoxelCodeBench (2026) evaluates 3D spatial reasoning through code generation: from minimal API documentation and a few examples, a model writes Python that builds voxel scenes in an Unreal Engine renderer, and human annotators judge the rendered output. This board is shape correctness in percent, the share of the 220 tasks whose output matches the specified geometry; higher is better.

Metric: Shape correctness (%): share of the 220 tasks whose rendered voxel output a human annotator judges to match the specified geometry (the paper's primary metric); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-587.9#178
2Claude Sonnet 4.580.4#261
3GPT-5 Mini80.4#364
4GPT-5 Chat69.7#449
5Claude Opus 469.4#326
6Claude 3.5 Sonnet66.9#752
7Claude 3 Opus45.2#1431
8Gemini 3 Pro19.5#145

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=voxelcodebench · How It Works · Data refreshed daily, snapshot 2026-11-21.