OpenAI strengthens AI models' ability to migrate legacy COBOL software

OpenAI says its models have become better at building and migrating software written in legacy programming languages such as COBOL, Fortran and Visual Basic, which have a shrinking pool of skilled programmers. According to the company, 40% of the world's banks still depend on COBOL systems, and migrating to modern languages could cost millions to hundreds of millions of dollars and take considerable time. As COBOL programmers age out of the workforce, banks face pressure to migrate in the coming years.

OpenAI created MultiLangBench, an evaluation suite that requires models to build SQLite-compatible database engines in a specified language. The company reported that GPT-6 Astra achieved an average pass rate of 87.4% across 20 languages, compared with 26.8% for GPT-5.6 Sol. COBOL remained the hardest language, with Astra scoring 72%.

OpenAI also tested whether its models could migrate COBOL programs to C++, Java, Kotlin, Python or Rust while preserving original behavior, such as rounding rules and handling of missing records. In these tests, GPT-6.1 Sol reached 97.9% accuracy with high reasoning effort and 86.6% with low reasoning effort. OpenAI noted that a migration is successful only if all tests pass, and that behavior-preserving capabilities are useful for maintaining and modernizing older systems, though production use still requires expert review and testing against business rules and dependencies.