Benchmark Finds Frontier AI Models Scoring as Low as 2% on Construction Drawing Takeoff
Togal.AI's benchmark indicates general-purpose AI models struggle to interpret construction drawings and produce estimating measurements, with the lowest score at 2%.
A benchmark published by construction-AI vendor Togal.AI found that frontier AI models perform poorly at interpreting construction drawings and producing measurements for estimating, with scores ranging from 2% to 30%. The findings circulated via Construction Briefing's technology listing on October 13 and remained visible in the portal's listings through late October.
The result supports a growing position in the construction technology sector: generic AI models are not reliable for technical construction documentation, and domain-specific pipelines remain necessary for takeoff and estimating workflows.
Caveats apply. The benchmark comes from a vendor with a commercial interest in the conclusion, the full article sits behind a bot wall, and the figures have not been independently replicated. Still, the score range — with the best frontier models under one-third accuracy — is a striking data point for anyone assuming general-purpose AI can read construction drawings out of the box.