OpenAI Enhances GPT-6 for Legacy Programming Languages Like COBOL and Fortran

OpenAI has enhanced its AI models for low-resource programming languages, including COBOL, Fortran, and Visual Basic. These languages, which have significantly less training data compared to popular languages like Python, are crucial for global infrastructure. According to IBM, COBOL supports over 40% of online banking systems and 95% of ATM transactions.

To assess its progress, OpenAI developed two evaluations. MultiLangBench challenges models to create a SQLite-compatible database engine from scratch. With maximum reasoning effort, GPT-6 Astra achieves an average test pass rate of 87.4% across 20 languages, a substantial increase from approximately 38% for GPT-5.6 Terra. Meanwhile, GPT-6 Luna scores 40.3%, slightly surpassing GPT-5.6 Terra’s 37.9% but at a lower cost.

On the COBOL migration benchmark, which translates COBOL into languages like C++, Java, Kotlin, Python, or Rust, GPT-6.1 Sol reaches 86.6% accuracy with low reasoning effort, compared to just 5.4% for GPT-5.6 Sol. When reasoning effort is increased, accuracy rises to 97.9%.

OpenAI’s advancements in these legacy programming languages could play a significant role in modernizing critical financial infrastructure.

Source: The Decoder