OpenAI says its latest model is now better at writing and migrating software in COBOL, Fortran, and Visual Basic.
An estimated 800 billion lines of COBOL remain in production. That code powers systems at banks, government agencies, and large enterprises.
What OpenAI announced
OpenAI shared the update on X on October 19, 2026. The company said its newest model improvements help with building and migrating code in what it calls low-resource languages.
In AI terms, “low-resource” describes a language with relatively little code for a model to learn from.
OpenAI framed the work around two jobs. The first is helping teams understand business logic that was written decades ago. The second is preserving how existing systems behave while they get modernized.
OpenAI is also asking developers and enterprise customers for input. It wants to hear which other languages or workflows could benefit from similar improvements.
Why legacy code is so hard for AI
COBOL has historically been a weak spot for coding models. Research benchmarks have shown low Pass@1 and compilation rates on COBOL tasks. Those rates have ranged from single digits to about 40%.
Pass@1 measures how often a model gets a working answer on its first try.
Legacy systems often use domain-specific dialects. They also carry business rules that may not be documented anywhere outside the code itself.
Fortran shows where things could be heading. Agentic AI workflows built on OpenAI models can reportedly translate and optimize Fortran code on their own. The cost can be as low as a few USD per kernel.
A crowded race to modernize old systems
OpenAI is not alone in chasing this market. Anthropic released COBOL-related updates in February 2026. IBM offers watsonx Code Assistant, a tool built to translate COBOL into Java.
IBM shares dropped by approximately 13% after Anthropic’s COBOL updates.
The workforce problem behind the push
Organizations running legacy systems face two pressures at once. Maintenance costs keep rising. The pool of engineers who know these languages keeps shrinking.
Successful modernization usually takes both AI tools and human oversight. Verification is the critical piece.
What this means for enterprises and the tech sector
In regulated industries, a migration that changes behavior even slightly can trigger compliance headaches. Any AI tool in this space will be judged on how reliably it keeps outputs identical, not just on whether the new code compiles.
OpenAI has not published comparable benchmark figures for its latest update in this announcement, so independent testing will matter.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.