OpenAI 称 GPT-6 显著提升 COBOL、Fortran 等老旧编程语言能力

OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran

文章要点

1OpenAI 称 GPT-6 系列对 COBOL、Fortran 等低资源编程语言能力显著提升
2GPT-6 Astra 官方 20 语言评测平均通过率 87.4%，GPT-5.6 Terra 约 38%
3COBOL 迁移基准上 GPT-6.1 Sol 准确率 86.6%，提高推理力度可达 97.9%
4IBM 估计 COBOL 支撑超 40% 网上银行系统和 95% 的 ATM 交易

决策要点

大模型开始补上老旧企业系统这块短板，COBOL 等遗留代码的迁移与维护从纯人力问题转向模型能力竞争。

影响：依赖 COBOL 等遗留系统的金融和保险企业多了一条现代化路径，旧系统改造预算需按新能力重算；基准成绩也可作为模型选型与供应商谈判的参考；目前数字均为官方自测。
参考：可考虑选一个非核心的 COBOL 模块做小规模翻译试点，比对人工改造成本；值得先盘点现有遗留系统的语言构成和规模，再算改造预算。
依据：全部成绩来自 OpenAI 自建的 MultiLangBench 和 COBOL 迁移基准，官方口径暂无独立验证；银行占比数据为 IBM 估算。

正文

OpenAI improved its latest AI models for "low-resource" languages like COBOL, Fortran, and Visual Basic. These languages have far less available training code than something like Python, but according to OpenAI, they prop up critical infrastructure worldwide. IBM estimates that COBOL powers more than 40 percent of online banking systems and 95 percent of all ATM transactions.

OpenAI built two evaluations to measure progress. MultiLangBench (GitHub) asks models to build a SQLite-compatible database engine from scratch. With maximum reasoning effort, GPT-6 Astra hits an average test pass rate of 87.4 percent across 20 languages, up from about 38 percent for GPT-5.6 Terra. GPT-6 Luna scores 40.3 percent, barely edging out GPT-5.6 Terra (37.9 percent) but at a much lower cost.

Average test pass rate on MultiLangBench across 20 programming languages. GPT-6 Astra and GPT-6.1 Sol both reach about 87 percent, while older models like GPT-5.6 Terra top out at around 38 percent. | Image: OpenAI

On the COBOL migration benchmark, which translates COBOL into C++, Java, Kotlin, Python, or Rust, GPT-6.1 Sol reaches 86.6 percent accuracy at low reasoning effort, compared to 5.4 percent for GPT-5.6 Sol. Cranking up reasoning effort pushes accuracy to 97.9 percent.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.