OpenAI's new models lift COBOL migration accuracy from 5.4% to 97.9%
Advanced coding in low-resource programming languages
OpenAI's latest models sharply improve coding in low-resource languages like COBOL, Fortran, and Visual Basic, which still run much of the world's banking, weather forecasting, and reactor safety software. On a new MultiLangBench that asks models to build a SQLite-compatible engine in 20 languages, GPT-6 Astra hits 87.4% average test pass rate, up from 26.8%. On a COBOL-to-modern-language migration test, GPT-6.1 Sol reaches 97.9% accuracy at high reasoning effort, preserving byte-for-byte behavior.
COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions and 95% of ATM transactions.