Text-to-CAD Generation on DeepCAD (test) benchmark leaderboard · SOTA2 Research

Text-to-CAD Generation on DeepCAD (test)

1.12MCDTOOLCAD

Evaluation Results

Method	Links
TOOLCAD2026.04		1.12	—	—	—	1.51	63.9
TOOLCAD2026.04		1.26	—	—	—	1.84	61.8
TextCAD2026.04		1.97	—	—	—	2.25	—
HNC-CAD2026.04		3.08	—	—	—	3.45	—
SkexGen2026.04		3.85	—	—	—	4.59	—
GPT-4o2026.04		5.88	—	—	—	9.12	62.7
CAD-Llama-INS2025.05		10.53	84.72	1.54	3.59	—	—
DeepCAD2026.04		13.41	—	—	—	15.36	—
LLaMA32025.05		17.33	17.26	4.1	12.36	—	—
GPT-4o2026.04		19.84	—	—	—	20.49	48.4
Text2CAD2025.05		20.64	69.91	3.02	9.98	—	—
CAD-Translator2025.05		21.29	70.36	2.94	10.92	—	—
GPT-42025.05		25.62	20.03	3.33	18.09	—	—
Qwen-2.5-7B-Instruct2026.04		28.49	—	—	—	15.21	43.7
Qwen3-8B2026.04		29.43	—	—	—	16.13	41.2
Qwen3-235B-A22B2026.04		29.83	—	—	—	25.65	60.5
Qwen-2.5-7B-Instruct2026.04		30.77	—	—	—	22.74	39.2
GPT-3.52025.05		32.56	20.39	3.27	13.6	—	—
Mistral2025.05		32.79	13.12	4.71	18.42	—	—
Qwen3-8B2026.04		33.64	—	—	—	24.58	37.8
Qwen3-235B-A22B2026.04		37.18	—	—	—	30.25	51.3
Qwen2.5-72B-Instruct2026.04		48.29	—	—	—	45.17	43.4
Qwen2.5-72B-Instruct2026.04		50.73	—	—	—	52.94	36.1
Qwen3-8B2026.04		53.18	—	—	—	54.02	27.9
Qwen-2.5-7B-Instruct2026.04		61.39	—	—	—	62.88	23.4
Qwen3-8B2026.04		61.78	—	—	—	60.39	18.5
Qwen-2.5-7B-Instruct2026.04		68.41	—	—	—	70.32	14.9