IT之家 10 月 21 日消息,OpenAI 公司昨日(10 月 20 日)发布研究报告,称增强 GPT-6 Astra 等模型构建和迁移 COBOL、Fortran、Visual Basic 等“低资源编程语言”的能力。这类语言虽然公开代码和文档较少,却广泛支撑银行、ATM、天气预报和核安全分析等关键平台。
为衡量能力,OpenAI 推出两项实测:MultiLangBench 要求模型用 20 种语言从零构建兼容 SQLite 的数据库引擎;另一项实测则要求模型保持原有业务行为不变的情况下,将 COBOL 程序迁移至 C++、Java、Python、Rust 等语言。
结果表明,GPT-6 Astra 在 MultiLangBench 的 20 种语言中平均测试借助率达 87.4%,较 GPT-5.6 Sol 的 26.8% 大幅提高;其中 COBOL 从 10.9% 升至 72.1%,Fortran 从 21.4% 升至 83.5%,Visual Basic 从 27.5% 升至 89.4%。
成本更低的 GPT-6 Luna 平均借助率为 40.3%,估算 API 成本较 GPT-5.6 Sol 降低约 78%。IT之家附上相关图片如下:
在 COBOL 迁移实测中,GPT-6.1 Sol 在低推理强度下准确率达 86.6%,远高于 GPT-5.6 Sol 的 5.4%,成本降低 39%;在更高推理强度下,准确率达到 97.9%,接近实测上限。
OpenAI 表示,这项进展有助于企业维护老旧平台、修复缺陷并推进软件现代化。不过,公司也强调,将 AI 迁移结果用于生产环境仍需专家审查,并结合具体业务规则和依赖关系开展测试。
本文来源:https://www.ithome.com/1/015/633.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT House October 21st news, OpenAI released a research report yesterday (October 20th), saying that it enhances the ability of models such as GPT-6 Astra to build and migrate COBOL, Fortran, Visual Basic and other "low-resource programming languages". While these languages have relatively little open code and documentation, they broadly support key platforms such as banking, ATMs, weather forecasting, and nuclear safety analysis.
In order to measure capabilities, OpenAI launched two tests: MultiLangBench required the model to build a SQLite-compatible database engine from scratch in 20 languages; another test required the model to migrate COBOL programs to C + +, Java, Python, Rust and other languages while keeping the original business behavior unchanged.
The results showed that the average test utilization rate of GPT-6 Astra in 20 languages of MultiLangBench reached 87.4%, which was a significant increase from 26.8% of GPT-5.6 Sol; COBOL increased from 10.9% to 72.1%, Fortran increased from 21.4% to 83.5%, and Visual Basic increased from 27.5% to 89.4%.

The lower cost GPT-6 Luna has an average borrowing rate of 40.3%, and the estimated API cost is about 78% lower than GPT-5.6 Sol. The relevant pictures are attached to the IT House as follows:
In the COBOL migration test, the accuracy of GPT-6.1 Sol reached 86.6% at low inference strength, which was much higher than the 5.4% of GPT-5.6 Sol, and the cost was reduced by 39%. At higher inference strength, the accuracy reached 97.9%, which was close to the upper limit of the test.
OpenAI says this progress helps businesses maintain aging platforms, fix defects, and modernize their software. However, the company also emphasized that the use of AI migration results for production environments still requires expert review and testing in conjunction with specific business rules and dependencies.
This article is from: https://www.ithome.com/1/015/633.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.
