OpenAI upgrades its model for COBOL, Fortran, and other legacy code
OpenAI announced on October 19, 2026, that its latest model improvements help with building and migrating code in low-resource languages like COBOL, Fortran, and Visual Basic.
An estimated 800 billion lines of COBOL remain in production, powering systems at banks, government agencies, and large enterprises. OpenAI framed its work around two key jobs: helping teams understand business logic written decades ago and preserving existing system behavior during modernization.
Legacy code challenges
COBOL has historically been a weak spot for coding models, with research benchmarks showing low Pass@1 and compilation rates ranging from single digits to about 40%. Legacy systems often use domain-specific dialects and carry undocumented business rules within the code itself.
Fortran shows potential progress, as agentic AI workflows built on OpenAI models can reportedly translate and optimize Fortran code independently at costs as low as a few USD per kernel.
Market competition
OpenAI faces competition in this space, with Anthropic releasing COBOL-related updates in February 2026 and IBM offering watsonx Code Assistant to translate COBOL into Java. IBM shares dropped by approximately 13% following Anthropic's announcement.
OpenAI is also asking developers and enterprise customers for input. It wants to hear which other languages or workflows could benefit from similar improvements.
Enterprise implications
Organizations running legacy systems face rising maintenance costs while the pool of engineers familiar with these languages shrinks. In regulated industries, even slight behavioral changes during migration can trigger compliance issues, making reliable output preservation critical.
OpenAI has not published benchmark figures for its latest update, leaving independent testing as the next step to verify its capabilities in this specialized domain.
Related
