IT之家 October 21 news: OpenAI yesterday (October 20) released a research report,stating that it enhanced the ability of models such as GPT-6 Astra to build and migrate "low-resource programming languages" like COBOL, Fortran, and Visual Basic.Although such languages have relatively little public code and documentation, they widely support critical systems such as banking, ATMs, weather forecasting, and nuclear safety analysis.
To measure the capability, OpenAI introduced two benchmarks: MultiLangBench requires models to build SQLite-compatible database engines from scratch in 20 languages; the other benchmark requires models to migrate COBOL programs to C++, Java, Python, Rust, and other languages while keeping the original business behavior unchanged.
Results show that GPT-6 Astra achieved an average test pass rate of 87.4% across the 20 languages in MultiLangBench, a major improvement over GPT-5.6 Sol's 26.8%; among them, COBOL rose from 10.9% to 72.1%, Fortran from 21.4% to 83.5%, and Visual Basic from 27.5% to 89.4%.
The more cost-effective GPT-6 Luna had an average pass rate of 40.3%, with an estimated API cost about 78% lower than GPT-5.6 Sol. IT之家 includes the relevant images below:
In the COBOL migration benchmark, GPT-6.1 Sol achieved an accuracy of 86.6% at low reasoning effort, far above GPT-5.6 Sol's 5.4%, with costs reduced by 39%; at higher reasoning effort, accuracy reached 97.9%, approaching the benchmark ceiling.
OpenAI said this progress helps enterprises maintain legacy systems, fix defects, and advance software modernization. However, the company also emphasized that using AI migration results in production environments still requires expert review and testing against specific business rules and dependencies.