ConTech · Oct 13
ModerateAI flunks a real takeoff test, scoring as low as 2 percent
A software firm called Togal built a hard test for AI. It had 33 takeoff jobs that need the AI to read plans and measure parts. The best known AI models, like those from OpenAI and Google, scored just 2 to 34 out of 100. The models got raw PDF drawings and had to count and measure on their own. Togal sells its own takeoff AI, so keep that in mind. Still, the test shows plain chat AI is not ready to do your takeoffs.
Why it matters to GCs
A wrong count in a takeoff can blow your bid. This is proof that a general chat tool will miss parts on a real plan. Use AI for a first pass if you want, but a person must check every number before it goes out.
Construction Briefing·Via POLR AI Hub