OpenAI says its models are now better at building and migrating software written in old programming languages, a skill that could matter to banks still running on mainframe code. The company points to COBOL, Fortran and Visual Basic as languages where the pool of working programmers is shrinking.
Why do banks still depend on COBOL?
The company said about 40 percent of the world’s banks still rely on COBOL systems. Moving those systems to newer languages can cost anywhere from millions to hundreds of millions of dollars, and it takes a long time. But the people who know COBOL are aging out of the workforce, which means banks that still depend on it will have to make the switch within the next few years.
OpenAI said its work on improving COBOL capabilities was guided by feedback from developers and enterprise customers. The company acknowledged that coding models have struggled with older languages, and it built a new evaluation suite to measure the gap. The suite, called MultiLangBench, asks models to create SQLite-compatible database engines in a specified programming language.
On that test, OpenAI said GPT-6 Astra posted an average pass rate of 87.4 percent across 20 languages. GPT-5.6 Sol scored 26.8 percent. COBOL remained the hardest language for the models, with Astra passing 72 percent of COBOL tests.
The company also tested whether models could translate COBOL programs into C++, Java, Kotlin, Python or Rust while keeping the original programs’ behavior intact. That matters in banking, where small details can change the outcome. OpenAI cited examples such as how a program rounds numbers with extra decimal digits and how it handles missing records.
How accurate are AI models at migrating COBOL code?
In that migration test, GPT-6.1 Sol achieved the highest accuracy, at 97.9 percent with high reasoning effort. At low reasoning effort, the same model reached 86.6 percent.
Does AI-assisted code migration still need human review?
OpenAI acknowledged in its post that a migration counts as successful only if “all tests pass.” Banking software, the company noted, has to be highly accurate. It also cautioned that the capabilities are meant to help teams maintain and modernize older systems. “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,” the company wrote.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.