Tyler Cowen's o1 Test: AI That Actually Reasons Economics

Economist Tyler Cowen just kicked the tires on OpenAI’s new o1 model series, and his initial verdict should make the econ profession sit up straight. Unlike chatbots that pattern-match their way to plausible-sounding nonsense, these models are built to pause and reason through complex problems before spitting out an answer. Cowen’s test was brutally simple: throw any economics question at it and see what sticks. The result? He found its performance not just good, but genuinely significant.

“The mere fact that you can ask it — as far as I can tell — any economics question and it has a good answer is really quite significant,” Cowen said. That’s not faint praise coming from a guy who has spent decades thinking about thinking. The o1 series represents a deliberate shift away from the instant-gratification paradigm of earlier large language models. OpenAI designed these systems to spend more time computing internally, a process that mimics the kind of deliberate, multi-step reasoning that separates a student who memorizes from one who actually understands general equilibrium.

What makes this different from past models is the reasoning architecture itself. Previous generations could ace a multiple-choice test by recognizing patterns but would crumble when asked to explain the intuition behind the answer. Early indications suggest o1 handles the “why” far better—a crucial distinction for a field like economics where policy arguments hinge on causal chains, not just correlations. Cowen’s broad assessment implies the model can navigate everything from undergraduate supply-and-demand questions to niche theories without hallucinating into absurdity. For economists who’ve spent years rolling their eyes at AI’s confident but incorrect takes on comparative advantage, that’s a low bar that surprisingly few models have cleared.

Still, the most interesting implication isn’t that AI can now pass a grad school qualifier. It’s what this means for the marginal product of human economists. If a reasoning engine can instantly provide a solid first draft of any economic argument, the value of rote knowledge drops to near zero. The premium shifts entirely to asking genuinely novel questions—and knowing which answers actually matter. Cowen seems to get this. He’s not marveling at a toy; he’s testing a tool that reshapes the workflow of his entire discipline.