OpenAI details coding gains in COBOL, Fortran and other less common languages

OpenAI said its newer models perform better on coding tasks in languages with relatively scarce training material. Its October 19 research update examines software development and migration involving languages including COBOL, Fortran and Visual Basic.

One evaluation, MultiLangBench, requires models to construct database engines compatible with SQLite in 20 languages. At maximum reasoning effort, OpenAI measured an average passing rate of 87.4% for GPT-6 Astra, compared with 26.8% for GPT-5.6 Sol.

A separate evaluation checks whether models can translate COBOL programs without changing their behavior. OpenAI said GPT-6.1 Sol achieved 86.6% accuracy at low reasoning effort, versus 5.4% for GPT-5.6 Sol, at 39% lower API cost.

These languages remain embedded in important business and scientific systems, where undocumented behavior can complicate maintenance. A migration can produce working software while still changing calculations or other business rules.

OpenAI said production migrations still require specialist review and tests covering the original system’s rules and dependencies. It plans to use the evaluations and developer feedback to guide further improvements.