New benchmark exposes AI’s difficulty with reading construction blueprints

Digital Journal explores how Construction’s Last Exam (CLE), developed by Togal.AI, reveals the challenges AI models face when interpreting construction blueprints. With leading models scoring just 2% to 34% on 33 real-world tasks, the benchmark highlights the gap between general AI capabilities and the precision required for construction takeoffs. The article also examines the importance of specialised AI tools, data security, and human oversight in bringing AI into construction workflows.