OpenAI’s New Models Improve Handling of Legacy Programming Languages
New models show major improvements in migrating legacy code to modern platforms.
OpenAI has announced an update to its latest artificial intelligence models, focusing on so-called "low-resource" programming languages. The update targets classic languages like COBOL, Fortran, and Visual Basic. While these are less popular than Python in the modern tech world, their importance to global infrastructure remains critical. Statistics show that COBOL underpins 40 per cent of banking systems and 95 per cent of ATM transactions. ## Testing and New Capabilities To evaluate progress, OpenAI created two special assessment methods. One, MultiLangBench, tasks models with creating a SQLite-compatible database engine. The results are impressive: GPT-6 Astra recorded an average success rate of 87.4 per cent across 20 different languages, a significant jump compared to the 38 per cent result of the previous generation, GPT-5.6 Terra. Simultaneously, the company introduced a more accessible version, GPT-6 Luna, which slightly outperformed its predecessor with a 40.3 per cent result while operating at much lower costs. ## Updating Critical Infrastructure The COBOL migration test is particularly noteworthy, as it required models to translate code into C++, Java, Kotlin, Python, or Rust. GPT-6.1 Sol showed 86.6 per cent accuracy in low-intensity reasoning mode, whereas GPT-5.6 Sol achieved only 5.4 per cent. By increasing reasoning resources, the new model's efficiency rises to 97.9 per cent. This update is vital for organisations migrating legacy systems to modern platforms. Automated rewriting of outdated code not only saves time but also reduces the likelihood of human error, which remains a major challenge for the financial and government sectors. For the Georgian market, this news is significant as local IT companies serving the large corporate sector will be able to manage technological legacy more effectively. The ability of AI to understand and transform decades-old code will significantly accelerate the process of paying down technical debt.