OpenAI Reports Major GPT-6 Coding Gains in COBOL and Fortran

OpenAI reports in a company blog post that its latest models perform better at coding in languages with limited training data, including COBOL, Fortran and Visual Basic. The company created two evaluations covering the construction of new software and the migration of existing COBOL programs.

In MultiLangBench, models must build a SQLite compatible database engine in 20 programming languages. GPT-6 Astra recorded an average test pass rate of 87.4%, compared with 26.8% for GPT-5.6 Sol. Its scores included 72.1% for COBOL, 83.5% for Fortran and 89.4% for Visual Basic. GPT-6 Luna reached 40.3% at an estimated average API cost of $0.77 per sample, about 78% below GPT-5.6 Sol.

A separate evaluation tested migrations from COBOL to C++, Java, Kotlin, Python or Rust while requiring identical observable outputs. GPT-6.1 Sol reached 86.6% accuracy at low reasoning effort, up from 5.4% for GPT-5.6 Sol, while costing 39% less. Accuracy rose to 97.9% with more reasoning. OpenAI also released the MultiLangBench setup and test suite on GitHub, and said migrations of production software still require expert review and testing.