OpenAI 称其模型可帮助银行从老旧 COBOL 程序迁移
OpenAI says its models can help banks migrate from old COBOL programs - Engadget
Engadget
以下为原文逐段中文翻译，保留图表与说明。AI 翻译并经 AI 对照复核，仍可能存在错误；核对数据和引用时请同时查看原文。译文发布：2026年10月20日 22:50（北京时间）。
该公司（OpenAI）的模型现在更擅长处理 COBOL 及其他老旧语言。
Casimiro PT/Shutterstock
OpenAI 的模型如今更擅长用日益减少的程序员群体所掌握的老旧编程语言（包括 COBOL、Fortran 和 Visual Basic）来构建和迁移软件。据该公司解释，全球 40% 的银行仍依赖 COBOL 系统，而要迁移到使用更新语言系统，不仅要花费数百万至数亿美元，还要耗费大量时间。然而，从事 COBOL 开发的程序员正逐渐退出劳动力市场，因此仍在使用该语言的银行将不得不在未来几年内完成迁移。
该公司表示，其改进模型 COBOL 能力的工作，由提交反馈的开发者和企业客户引导。在应对老旧语言方面，编程模型显然仍很吃力。OpenAI 创建了一个名为 MultiLangBench 的 AI 基准测试（benchmark）套件，要求模型用指定编程语言创建兼容 SQLite 的数据库引擎。据称 GPT‑6 Astra 在 20 种语言上的平均测试通过率为 87.4%，比通过率仅 26.8% 的 GPT-5.6 Sol 高效得多。COBOL 仍然是最难攻克的编程语言，Astra 的通过率为 72%。
OpenAI 还让模型将 COBOL 程序迁移到使用 C++、Java、Kotlin、Python 或 Rust 的程序，同时保留程序原有功能，以此评估其模型能力。要让 OpenAI 的模型有效帮助银行机构迁移到现代语言，它们必须能够保留 COBOL 程序的行为，例如如何处理多余小数位的四舍五入，以及如何应对缺失记录。在这种情况下，GPT-6.1 Sol 在高推理强度下达到了最高精度，为 97.9%。在低推理强度下，其精度为 86.6%。
OpenAI 在帖文中承认，只有「所有测试通过」，迁移才算成功。毕竟，处理银行任务的程序需要极高的准确性。「这类行为保留能力对维护和现代化改造老旧系统的团队很有用，不过将其应用于生产软件，仍需要专家审查，并针对系统业务规则与依赖项进行测试，」文中写道。
展开英文原文 · 与中文逐段对应
OpenAI says its models can help banks migrate from old COBOL programs - Engadget
The company’s models are now better at working with COBOL and other old languages.
Casimiro PT/Shutterstock
OpenAI's models are now better at building and migrating software in old programming languages with a dwindling number of programmers, including COBOL, Fortran and Visual Basic. As the company explains , 40 percent of world's banks still rely on COBOL systems, and migrating to ones that use newer languages would not only cost them millions to hundreds of millions, but also a lot of time. Programmers that work with COBOL are aging out of the workforce , however, so the banks that still use it will have to migrate within the next few years.
The company said its work on improving its models' COBOL capabilities is guided by developers and enterprise customers who submitted feedback. Coding models are apparently struggling when it comes to working with old languages. OpenAI created an AI evaluation suite called MultiLangBench, which asks models to create SQLite -compatible database engines in a specified programming language. It said GPT‑6 Astra had an average test pass rate of 87.4 percent across 20 languages, making it a lot of effective than GPT-5.6 Sol that had a 26.8 percent pass rate. COBOL remains the hardest language to crack, with Astra's pass rate being 72 percent.
OpenAI also evaluated its models' capabilities by having them migrate COBOL programs to ones that use C++, Java, Kotlin, Python or Rust, while also preserving what the programs do. For OpenAI's models to effectively help banking institutions migrate to modern languages, they must be able to preserve COBOL programs' behaviors, such as how it rounds numbers with excess decimal digits and how it handles missing records. In this case, GPT-6.1 Sol achieved the highest accuracy, which reached 97.9 percent, with high reasoning effort. At low reasoning effort, it achieved 86.6 percent accuracy.
In its post, OpenAI admits that a migration is only successful if "all tests pass." Programs that handle banking tasks, after all, would need to be highly accurate. "Behavior-preserving capabilities like these are useful for teams maintaining and modernizing older systems, though applying it to production software still requires expert review and testing against the system's business rules and dependencies," it wrote.
查看本期独立摘要
中文独立摘要 · AI 整理
这件事
OpenAI称其模型对COBOL、Fortran等老旧语言的构建与迁移能力显著提升。全球约40%银行仍依赖COBOL，而相关程序员正退出劳动力市场。OpenAI推出MultiLangBench基准，GPT-6 Astra在20种语言平均通过率87.4%；迁移需保留原有行为，GPT-6.1 Sol高推理强度下精度达97.9%。
本页概括所列来源的重点，属于独立摘要；完整报道请阅读原站。整理时间：2026年10月20日 22:57（北京时间）。
信息从哪里来
据Engadget报道，引述OpenAI官方说明与MultiLangBench数据。
- Engadget 报道正文