Text-to-CAD generation on CADBench-Sim human evaluation
0.7693Attribute FidelityGraph-CAD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Graph-CADBackbone=Llama3-8B, Training Strategy=SAPCL2026.04 | 0.7693 | 0.7509 | 0.5482 | 0.6894 | |
| Claude-opus-4-1Paradigm=Graph-CAD2026.04 | 0.7521 | 0.7434 | 0.4962 | 0.6639 | |
| GPT-5Paradigm=Graph-CAD2026.04 | 0.7351 | 0.7237 | 0.4398 | 0.6328 | |
| Graph-CADBackbone=Llama3-8B, Training Strategy=SFT2026.04 | 0.7308 | 0.7326 | 0.4753 | 0.6462 | |
| Claude-opus-4-1Paradigm=end-to-end2026.04 | 0.7185 | 0.7298 | 0.546 | 0.4458 | |
| GPT-5Paradigm=end-to-end2026.04 | 0.7146 | 0.7281 | 0.4507 | 0.6311 | |
| BlenderLLM2026.04 | 0.6914 | 0.6862 | 0.3759 | 0.5845 |