OpenAI says its models can help banks move off COBOL

The company said 40 percent of the world's banks still rely on COBOL systems. Migrating those systems to newer languages would cost millions to hundreds of millions and take significant time, OpenAI said, but banks that remain on COBOL will have to move within the next few years because COBOL programmers are aging out of the workforce.

OpenAI said its work to improve the models' COBOL capabilities was guided by feedback from developers and enterprise customers. Coding models have apparently struggled with older languages, the company said. To measure progress, OpenAI created an evaluation suite called MultiLangBench, which asks models to create SQLite-compatible database engines in a specified programming language.

On that benchmark, GPT-6 Astra had an average test pass rate of 87.4 percent across 20 languages, compared with 26.8 percent for GPT-5.6 Sol, OpenAI said. COBOL remained the hardest language to crack, with Astra's pass rate at 72 percent.

OpenAI also tested whether its models could migrate COBOL programs to C++, Java, Kotlin, Python or Rust while preserving what the programs do. For banking institutions, that means preserving behavior such as how a program rounds numbers with excess decimal digits and how it handles missing records, the company said.

GPT-6.1 Sol achieved the highest accuracy in that migration test, reaching 97.9 percent with high reasoning effort and 86.6 percent with low reasoning effort, according to OpenAI.

OpenAI acknowledged that a migration is successful only if "all tests pass," and that banking programs need high accuracy. It said behavior-preserving capabilities are useful for teams maintaining and modernizing older systems, but applying them to production software still requires expert review and testing against the system's business rules and dependencies.