2026年10月6日，美国人工智能公司OpenAI宣布，其搭载未公开大模型的AI智能体在短短数周内攻克了至少372个长期未解的数学难题，并在超过700篇AI生成的学术论文中详细公布了相关成果。此前一个月，该公司刚声称该模型仅耗时88小时便解决了“千禧年大奖难题”之一的纳维-斯托克斯方程问题。此次攻克的难题涵盖代数、几何、概率论及数理逻辑等诸多领域，AI智能体解出每个问题所需的平均算力仅相当于其公开发布模型约3小时的推理计算量。
在AI给出的海量成果中，还包含了对黎曼猜想的推进尝试以及对棘手的“三维挂谷极大猜想”（3D Kakeya maximal conjecture）的证明声明——纽约大学柯朗研究所的王虹曾因在该领域的奠基性研究于今年7月荣获菲尔兹奖。然而，这372项解法仅仅是OpenAI向智能体投喂的4000个难题中的一部分，公司并未公开未获突破的题目细节。此外，许多论文篇幅超过100页，其中仅有127篇配备了数学界通行的形式化逻辑验证代码，且仅公开了10个难题的完整内部推理链条，使学术界全面审阅面临巨大挑战。
这种“批量化制造数学答案”的模式引发了学术界的深刻警惕。在纳维-斯托克斯声明后不久，28位菲尔兹奖得主发表联名公开信，警告将未解数学难题单纯作为大模型基准测试工具可能扼杀深度理论探索与方法创新，该公开信迄今已获得逾8000名数学同行的联署。作为回应，OpenAI参考了一个由9位著名数学家组成的独立顾问小组的建议，承诺资助一系列国际研讨会并公开部分推理过程，以协助人类学者消化这一前所未有的知识洪流并厘清AI对数学本质的深远冲击。
On October 6th 2026, OpenAI announced that autonomous AI agents running an unreleased model had resolved at least 372 long-standing mathematical problems within weeks, detailing the proofs across more than 700 machine-generated papers. This announcement followed claims last month that the same internal model cracked the Millennium Prize Navier-Stokes problem in just 88 hours. Covering domains from abstract algebra to logic, the autonomous agents derived solutions requiring an average compute equivalent to approximately three hours of reasoning on the firm's commercially available public models.
The release highlighted claimed proofs concerning the Riemann hypothesis as well as the 3D Kakeya maximal conjecture, a problem intimately connected to work for which NYU Courant mathematician Hong Wang received the Fields Medal in July. However, these 372 solutions derived from a larger pool of 4,000 problems submitted to the models, with OpenAI providing no performance diagnostics on unsolved prompts. Furthermore, with numerous individual papers exceeding 100 pages, only 127 problem solutions included machine-checkable formal code ensuring logical consistency, and the agents’ internal reasoning traces were released for only ten cases.
This industrialized approach to mathematical theorem proving has ignited intense pushback across academia. Following the Navier-Stokes claim, 28 Fields Medallists issued an open letter arguing that utilizing open conjectures as benchmarks risked terminating fruitful theoretical inquiries without generating foundational human insight, an initiative subsequently endorsed by more than 8,000 professional mathematicians. In response, OpenAI convened an independent advisory panel of nine mathematicians, pledging conference sponsorships and limited reasoning releases to assist researchers in evaluating whether automated computation expands mathematical epistemology or merely proliferates mechanical verification.
Source: OpenAI claims to have solved hundreds of long-standing maths problems
Subtitle: It is judgment day for mathematics
Dateline: Oct 8th 2026\n