Recently, while browsing NEAR’s GitHub, I came across something interesting: The NEAR AI team has open-sourced a test in which they successfully solved 672 extremely challenging mathematical proof problems using the Chinese open-source large model DeepSeek. After asking the AI, I learned these problems are called PutnamBench—some of the most difficult questions from North America’s premier undergraduate mathematics competition. In simple terms, passing this test demonstrates that an AI possesses highly rigorous logical reasoning capabilities. According to @MSBIntel, NEAR’s solution completed all 672 problems at an average cost of just $0.17 per question, totaling slightly over $110. By comparison, other solutions achieving perfect scores used OpenAI models at an average cost of $44.50 per question, while Claude required $3.52 per question. Why would a blockchain focus on such advanced mathematical problems? The real goal behind investing significant effort into having AI solve mathematical proofs is smart contract security auditing—technically known as formal verification. For example, using mathematical logic to inspect code line by line for vulnerabilities; if low-cost, open-source AI compute power can reduce the cost of verification to just a few dimes— in the future, even projects and individual developers might affordably secure and audit their contracts against safety failures. While this has no direct link to the price of near:native, it does signal that the project team is actively building meaningful technology.