Construction & Engineering
AI models struggle with construction drawing analysis
Togal.AI says leading AI models scored only 2% to 34% on a 33-task benchmark for reading construction drawings and calculating quantities.
An assessment by Togal.AI tested general-purpose AI models on construction drawings, including tasks involving element identification and quantity calculations used in estimating. The benchmark covered 33 construction-related tasks.
The models scored between 2% and 34%, according to the company. The results indicate that general-purpose AI still struggles with technical drawing interpretation and the precision needed for construction estimating.
Drawing analysis and quantity takeoff are important because errors can affect bids, procurement, cost plans, and project schedules. Low scores suggest that firms should treat automated outputs as assistive rather than fully reliable without professional review.
The benchmark provides a practical measure of the gap between broad AI capabilities and specialized construction requirements. It also points to a likely need for domain-specific training data, structured drawing formats, validation systems, and human oversight before these tools can be trusted in high-stakes estimating workflows.
Original source: Bricks & Bytes