VoxelBench
Language models turn text or image prompts into voxel structures. Voters compare anonymous builds from the same prompt, and Glicko-2 ratings rank the models.
- GPT-6.1 Sol 2,607
- GPT-6 Astra 2,563
- Claude Opus 5.5 2,521
- GPT-6 Sol 2,361
- GPT-6.1 Terra 2,341
Official leaderboard