Analysis Two AI Takeoff Benchmarks Landed Eleven Weeks Apart. They Don't Agree on How Good the Models Are. Togal.AI has published Construction's Last Exam, 33 quantity-takeoff tasks on real drawings, with a public per-model scoreboard. The top general-purpose model on it scores 34.0; two score under 2. A rival takeoff benchmark from Handoff, published in July, puts frontier models in the 40s and 50s — and the disagreement is more informative than either number. Hugo Pegley · Oct 19, 2026
