OpenAI highlights low-resource coding

OpenAI says its latest models are getting better at coding in languages with less training material, including COBOL, Fortran, and Visual Basic. Those languages still support important business and public services, but maintaining their software can be hard when documentation is incomplete or the people who built it have left.

OpenAI tested two skills: building a SQLite-compatible database engine in 20 languages, and migrating COBOL programs while preserving their behavior. In the first evaluation, GPT-6 Astra passed an average of 87.4% of tests, compared with 26.8% for GPT-5.6 Sol. In the COBOL migration test, GPT-6.1 Sol reached 86.6% accuracy at low reasoning effort, rising to 97.9% at higher effort. These are OpenAI’s evaluation results, not a guarantee that a model can safely change a company’s live systems. OpenAI says production use still needs expert review and testing against business rules and dependencies. For an operator, the practical takeaway is that AI may help teams work through older code, but any proposed changes need careful checks before they affect customers or operations. Read OpenAI’s evaluation details.