A legendary maths riddle finally has an answer. OpenAI says one of its general-purpose AI models has autonomously solved a famous problem first posed in 1946, with external mathematicians checking the proof. The company calls it a major milestone for AI reasoning and scientific research.

The breakthrough concerns the planar unit distance problem, attributed to Paul Erdős. Put simply, if you scatter n points on a flat surface, how many pairs can be exactly one unit apart?

For nearly 80 years, the best-known constructions looked a lot like square grids. OpenAI reports that its model has now overturned this belief by discovering a new family of arrangements that outperforms grid-like patterns.

OpenAI describes the system as a general-purpose reasoning model rather than a tool built only for maths. It handled long, complex chains of logic without step-by-step human guidance, and its proof was later checked by independent experts. This makes the advance notable not just for the answer, but for the way the answer was found.

The company argues this is the first time an AI system has autonomously solved a prominent open problem central to an active branch of mathematics. Until recently, AI headlines focused on text, images or code. Producing a durable mathematical proof demands sustained logical consistency, precise definitions and results that survive expert scrutiny. That is why many researchers see this as a turning point for AI’s ability to reason.

Although the unit distance problem sits in discrete geometry, its ideas map to real systems. Arranging points efficiently shows up in network design, chip layout, wireless communication, robotics and materials science.

Insights about how to pack or connect things with fixed distances can inform sensor grids, circuit topologies and even crystal structures. The new constructions unearthed by the model could therefore inspire fresh designs in these areas.

OpenAI frames the result as evidence that modern models can sustain long reasoning, connect tools from different areas and produce work strong enough for outside verification. If such systems continue to mature, they could become collaborators that propose ideas, test them quickly and help scientists navigate hard problems in biology, physics, engineering and medicine. The real story may not be one solved puzzle, but a step toward dependable AI partners in research.

一个传奇数学谜题终于有了答案。OpenAI 称，其一个通用 AI 模型自主解决了一个 1946 年首次提出的著名问题，外部数学家已验证了证明过程。该公司称之为 AI 推理与科学研究的重要里程碑。

该突破涉及“平面单位距离问题”，由数学家保罗·埃尔德什提出。简单来说：在一个平面上随机放置 n 个点，最多有多少对点之间的距离恰好等于 1？

近 80 年来，已知的最佳构造大多类似正方形网格。OpenAI 称，其模型推翻了这一认知，发现了一组新的点阵排列，其性能超越了网格类结构。

OpenAI 将该系统描述为一个通用推理模型，而非仅为数学打造的专用工具。它能处理长而复杂的逻辑链，无需人为引导。这项进展之所以引人注目，不仅是因为答案本身，更因为答案的发现方式。

这是 AI 系统首次自主解决一个活跃数学分支中的著名开放问题。此前，AI 的头条新闻多聚焦于文本、图像或代码。而要产出经得起推敲的数学证明，需要持续的逻辑一致性、精确的定义以及能经受住专家审视的结果。因此，许多研究者将这一事件视为 AI 推理能力的转折点。

虽然单位距离问题属于离散几何，但其思想可以映射到现实系统中。高效排列点集的方法会出现在网络设计、芯片布局、无线通信、机器人学以及材料科学等领域。如何以固定距离打包或连接物体，相关洞见能为传感器网络、电路拓扑乃至晶体结构提供启发。因此，AI 模型发现的新型构造有望在这些领域激发新的设计思路。

OpenAI 将这一成果视为现代模型能够维持长程推理、连接不同领域工具，并产出足以通过外部验证的成果的有力证据。如果这类系统持续成熟，它们有望成为科研中的合作者 —— 快速提出想法、检验假设，并帮助科学家应对生物学、物理学、工程学和医学中的难题。真正的意义或许不在于解答了一道难题，而在于朝着值得信赖的 AI 研究伙伴迈出了一步。

(Translated by DeepSeek)