OpenAI just dropped a new benchmark called GDP-Val that tests AI models on real economic tasks like nursing plans and engineering calculations instead of academic tests. The results show Claude actually beats ChatGPT, and experts working with AI can save significant time and costs—but only if the model pushes back with criticism instead of just agreeing with everything.