VoxelCodeBench - Geometric Construction: 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 on the 50 geometric construction tasks, procedural shapes built with boolean operations; higher is better.

Metric: Shape correctness (%) on the geometric construction (50 tasks: parameterized primitives and boolean composition) tasks, judged by a human annotator; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 4 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-566.7#178
2Claude Sonnet 4.552.8#261
3Claude Opus 440#326
4Gemini 3 Pro23.7#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-geometric-construction · How It Works · Data refreshed daily, snapshot 2026-11-21.