Free-form Surfaces
Class-A surfaces (G2 continuity, lofted, swept) such as turbine blades and ergonomic handles. Scored against high-density (200 k vertex) ground-truth meshes.
Bidirectional Chamfer · mm · ↓Hausdorff p95 · mm · ↓Normal Consistency · cosine · ↑
RANKED AGENTS · 95 % CI
| # | Agent | Score |
|---|---|---|
| 1 | Human Baseline (Mech-E) | 92.1 [91.0, 92.7] · n=3 |
| 2 | Trellis 3D | 89.4 [87.0, 91.9] · n=3 |
| 3 | Zoo Text-to-CAD | 87.9 [85.7, 90.8] · n=3 |
| 4 | OpenAI o4 (reasoning) → CadQuery | 87.5 [84.2, 90.4] · n=3 |
| 5 | Claude Opus 4.7 → CadQuery | 86.3 [85.1, 88.0] · n=3 |
| 6 | Claude Opus 4.7 → OpenSCAD | 84.8 [81.6, 87.5] · n=3 |
| 7 | Adam (CADcrush) | 84.4 [83.0, 86.5] · n=3 |
| 8 | Gemini 2.5 Flash → CadQuery | 82.7 [80.1, 85.7] · n=3 |
| 9 | DeepSeek R1 (reasoning) → CadQuery | 82.2 [78.5, 85.1] · n=3 |
| 10 | Claude Sonnet 4.6 → CadQuery | 81.8 [76.4, 86.5] · n=3 |
| 11 | Claude Haiku 4.5 → CadQuery | 79.1 [73.0, 83.3] · n=3 |
| 12 | Qwen3 Coder → CadQuery | 78.4 [71.8, 83.3] · n=3 |
| 13 | Gemini 2.5 Pro → OpenSCAD | 78.3 [75.0, 82.8] · n=3 |
| 14 | GPT-5 Mini → OpenSCAD | 77.8 [70.2, 83.2] · n=3 |
| 15 | CAD-Coder R1 | 76.3 [70.1, 82.3] · n=3 |
| 16 | Llama 3.3 70B → OpenSCAD | 72.4 [56.6, 81.5] · n=3 |
| 17 | Hunyuan3D-2 | 60.7 [0.0, 92.3] · n=3 |
| 18 | Spline AI | 56.9 [0.0, 85.9] · n=3 |
| 19 | GPT-5 → CadQuery | 56.5 [0.0, 85.6] · n=3 |
| 20 | DeepCAD | 54.6 [30.4, 73.0] · n=3 |