Test of Claude Opus 5.5 on architectural drafting: capability and limits
What happened
On November 3, Hacker News featured an architect's architectural drafting test of 13 AI models. The tester concluded that Claude Opus 5.5, under professional guidance, can produce permit-quality drawings for simple buildings. The model generates scalable vector lines via Python code that can be edited in CAD; a residential drawing took twelve hours and cost hundreds of dollars in compute. Each model got different amounts of human feedback, and Opus 5.5 got the most, including two rounds of redline comments. The final drawing still missed a second means of egress required by local code, showing the result depended on human guidance and still had compliance gaps.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageClaude Opus 5.5 Can Draw Permit-Level Architectural Drawings
建筑师测试13个AI模型后认为，Claude Opus 5.5在专业指导下可为简单建筑生成达到建筑许可质量的图纸。模型通过Python代码生成可缩放、可在CAD中编辑的矢量线条，住宅图纸耗时十二小时、计算成本数百美元。各模型获得的人工反馈量不同，Opus 5.5接受了最多反馈，包括两轮红线批注，最终图纸仍遗漏当地法规要求的第二条疏散路径。
Heat over time
Not enough continuous observations to draw a trend yet.