OpenAI Targets COBOL and Fortran to Overhaul Legacy Banking Code
OpenAI announced that its latest model improvements significantly bolster support for low-resource and legacy programming languages, specifically naming COBOL, Fortran, and Visual Basic. The update aims straight at the aging mainframe software quietly powering global banking and government infrastructure.
Tackling the code that quietly runs global finance
Financial systems and government registries still depend heavily on languages written decades ago.
Modern developer tooling rarely bothers with COBOL or Fortran. While foundation models have become proficient at writing Python, TypeScript, and Rust, older languages suffer from a severe data deficit: there are relatively few public GitHub repositories or modern tutorials to train on.
That neglect has created an expensive problem. Trillions of dollars in daily commerce, airline ticketing engines, and social security payments still execute on codebases conceived during the Cold War. Qualified engineers who understand these systems are retiring, and replacing them has turned into a slow, high-risk operational headache.
How OpenAI handles low-resource legacy languages
OpenAI says its recent architectural updates improve reasoning across low-resource programming environments. Rather than simply guessing syntax, the models are tuned to map out legacy procedural logic and translate it into maintainable modern codebases.
Competitors are racing down the same track. As cloud hyperscalers push enterprises to modernize infrastructure—a shift underscored by how AWS and Anthropic are deploying engineers to demonstrate clear software returns—automated refactoring has become a primary enterprise battleground.
The real test will be accuracy. Translating a business rule written in 1978 into a modern microservice is rarely a one-to-one mapping. Hidden side effects, mainframe-specific quirks, and undocumented patches mean enterprises cannot simply click migrate and walk away.
Rewriting legacy software is notoriously dangerous. Most banks don't leave their core ledgers in COBOL because they love 1960s syntax; they do it because failed rewrite attempts have cost institutions hundreds of millions of dollars. If OpenAI can reliably deconstruct business logic without altering arithmetic behavior, it will save enterprise IT departments billions. But human verification will remain non-negotiable.
Availability and enterprise access
OpenAI is rolling out these legacy coding capabilities across its latest enterprise tiers and API models. Organizations modernizing mission-critical environments will access the features through their standard platform agreements, with initial availability focused on enterprise developers in North America and Europe.
Refactoring decades-old mainframes has long been one of enterprise tech's most dreaded chores. By deliberately tuning its models for COBOL and Fortran, OpenAI is betting that modernizing archaic code will be one of generative software's clearest commercial wins.
Frequently asked questions
Why is COBOL still used by modern banks?
COBOL excels at high-volume, fixed-point financial arithmetic and has powered mainframe systems reliably for decades. Replacing systems with millions of lines of active code carries massive operational risk.
What makes Fortran and COBOL low-resource for AI models?
Unlike Python or JavaScript, legacy enterprise languages reside mostly in proprietary corporate mainframes rather than public open-source repositories, giving frontier models far fewer public tokens to learn from.
Sources
- OpenAI Modellerine COBOL ve Fortran Desteği: Eski Sistemleri Dönüştürecek. @OpenAI, 2026. https://x.com/OpenAI/status/2112310345863930168