AI Is Getting Good at Math. Here’s Why.
AI models from OpenAI and Anthropic are solving major open mathematical problems dating back decades.
“The models are suddenly good at math.”
OpenAI and Anthropic have recently solved multiple long-standing open mathematical problems, including a counterexample to Erdős's unit distance conjecture (open since 1946), proofs of Navier-Stokes and the Cycle Double Cover problem, and formalization of Fermat's Last Theorem. The article frames this capability leap through a complexity model: problem difficulty grows exponentially with the number of required insights, suggesting AI may be unlocking a new class of tractable problems. This represents a critical signal about where frontier AI capabilities are heading in formal reasoning and scientific discovery.