AI Models Used in Modernization of COBOL Code
OpenAI announced that it has achieved up to 98 percent accuracy in tests for converting legacy COBOL software to modern languages, while experts have issued warnings.
OpenAI has announced that it is enhancing the capabilities of AI models in migrating legacy COBOL-based software, used in critical areas such as banking and government, to modern languages like C++, Java, Kotlin, Python, and Rust. In internal tests, the GPT-6.1 Sol model achieved a 98 percent behavior preservation accuracy with a high level of reasoning.
This development, announced on October 19, aims to help organizations maintain and modernize their critical infrastructure. The model, which offers 87 percent accuracy with low reasoning effort, increased its success rate to 98 percent when the effort was increased. Meanwhile, Anthropic had previously made a similar market move with its 'Claude Code' agent.
Software engineering experts Vadim Zaytsev and Arie van Deursen emphasized that rewriting code does not mean changing the system entirely. The experts stated that breaking down multilingual and complex legacy systems into manageable parts and preventing errors, which are unacceptable in critical infrastructure, are the fundamental challenges. OpenAI spokesperson Oscar Haines also acknowledged that legacy code migration remains a complex and unresolved problem.
Sources
- theverge.com · Link