OpenAI has updated its model to improve performance in writing and migrating software using legacy languages, specifically COBOL, Fortran, and Visual Basic. The company announced these enhancements on X on October 19, 2026, focusing on low-resource languages where training data is limited.
The initiative aims to assist teams in deciphering decades-old business logic and maintaining system consistency during modernization. OpenAI is currently soliciting feedback from developers and enterprise clients to identify additional languages or workflows that require similar support.
Historically, COBOL has presented challenges for coding models, with benchmark Pass@1 and compilation rates ranging from single digits to 40%. While agentic AI workflows can now translate and optimize Fortran code for a cost as low as a few USD per kernel, independent testing remains necessary due to a lack of published benchmark figures.
The market for legacy system modernization is competitive, with Anthropic having released similar COBOL-related updates in February 2026 and IBM offering its watsonx Code Assistant for Java migration. Organizations are increasingly turning to AI tools to mitigate rising maintenance costs and a shrinking pool of specialized engineers.
