Togal.AI launches 'Construction's Last Exam' benchmark; general AI models score 2% to 34% on takeoff
Announced · Published · By Contractor AI Shop
Togal.AI released Construction's Last Exam, a 33-task benchmark that tests AI models on reading construction drawings and producing takeoff quantities. Construction Briefing reported general-purpose models scored between 2% and 34%. Togal sells takeoff software, so it is not a neutral party.
Key facts
- Togal.AI announced Construction's Last Exam on October 6, 2026. It covers 33 real-world tasks including reading plans and measuring spaces and materials. Togal.AI (via EIN Presswire)
- The benchmark site lists 12 area tasks, 12 counting tasks and 9 linear tasks. Togal.AI
- Construction Briefing reported general-purpose models scored between 2% and 34%, given raw PDFs and no third-party tools. Construction Briefing
- Construction Metrics reported Togal's own platform, run with a person in the loop, scored 97.7. Togal both built the test and sells the product. Construction Metrics
Our take: what it means for contractors
This backs up what many estimators already see: general chatbots aren't ready to do takeoff from PDFs on their own. Use ChatGPT or Claude to read specs and draft scope letters. For quantities, use a dedicated takeoff tool and check it against your own counts.
What we don't know yet
- Independent replication by someone without a product to sell
- How results change as new models are released
Sources
- Togal.AI (via EIN Presswire) · Togal.AI Launches 'Construction's Last Exam,' a Benchmark Testing AI on Real-World Construction Tasks · 2026-10-06 · Primary
- Togal.AI · Construction's Last Exam · Primary
- Construction Briefing · AI models score as low as 2% on construction takeoff test · 2026-10-13 · Secondary
- Construction Metrics · Togal.AI's takeoff benchmark scores the best general AI model at 34.0 out of 100 on 33 drawing tasks · 2026-10-08 · Secondary