CAD code generation on CADBench Wild
76.95Attribute ScoreGraph-CAD (SAPCL)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Graph-CAD (SAPCL)Backbone=Qwen3-8B2026.04 | 76.95 | 75.9 | 60.57 | 71.14 | 2.5 | 65.77 | 89.43 | |
| Graph-CAD (SFT)Backbone=Qwen3-8B2026.04 | 69.44 | 72.7 | 58.61 | 66.92 | 4.5 | 63.58 | 80.25 | |
| GPT-5inference_protocol=three-stage inference, reasoning_mode=true, shots=22026.04 | 68.58 | 70.91 | 55.95 | 65.15 | 5.5 | 60.03 | 40.17 | |
| Claude-opus-4-1inference_protocol=three-stage inference, reasoning_mode=true, shots=22026.04 | 68.47 | 72.18 | 59.97 | 66.87 | 14.5 | 55.5 | 50.62 | |
| BlenderLLM2026.04 | 67.82 | 63.63 | 45.81 | 59.09 | 5.3 | 60.56 | 49.83 | |
| Deepseek-r1inference_protocol=three-stage inference, reasoning_mode=true, shots=22026.04 | 48.14 | 51.41 | 37.35 | 45.64 | 20.5 | 48.58 | 42.75 | |
| Text2CAD2026.04 | 41.98 | 30.82 | 13.23 | 28.68 | 14 | 52.11 | — | |
| CADFusion2026.04 | 38.22 | 37.16 | 14.96 | 30.11 | 11.5 | 52.78 | — | |
| Llama-3.1-405binference_protocol=three-stage inference, shots=22026.04 | 33.31 | 35.3 | 19.43 | 29.34 | 47.2 | 32.42 | 29.03 | |
| Qwen-Plusinference_protocol=three-stage inference, reasoning_mode=true, shots=22026.04 | 25.96 | 27.22 | 19.51 | 24.23 | 61 | 24.46 | 13.05 | |
| Gemini-2.5-proinference_protocol=three-stage inference, reasoning_mode=true, shots=22026.04 | 20.02 | 18.8 | 16.67 | 18.5 | 48.7 | 17.5 | 25.84 |