OpenAI says its new models are better for building and migrating software in low-resource legacy programming languages like COBOL, Fortran, and Visual Basic
Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.
Context & Ripple Effects
OpenAI had already promoted GPT-5’s broad coding scores, while its own SWE-Lancer research found real freelance software tasks difficult for frontier models. The new claim narrows the coding pitch to languages with relatively little training material but large installed codebases.
It also arrives as model makers increasingly rely on internal benchmarks as public tests lose discriminating power. That makes evidence of legacy-language performance a product-positioning issue as much as a raw-model claim.
First-order effects
- Banks and government organizations maintaining COBOL, Fortran, or Visual Basic systems gain a new OpenAI option for code construction and migration work in those environments.
- OpenAI can position its coding models against tools optimized for mainstream languages by emphasizing support for legacy estates that remain costly to change.
Second-order effects
- IBM, whose Granite code models were trained across 116 languages, and Microsoft, which has introduced a GitHub-tuned coding model, face a higher bar to demonstrate useful performance on specialized enterprise code rather than broad coding benchmarks alone.
- Enterprise buyers evaluating AI-assisted migration will need task-level validation for their own codebases, since prior OpenAI testing found real-world software work more demanding than benchmark results imply.
Third-order effects
- If legacy-language results translate into dependable migration workflows, modernization work can shift from a scarce-language labor problem toward a model-assisted, governed software-delivery process.
- The industry’s evaluation center of gravity would move further from public coding tests toward proprietary benchmarks and customer-specific proof of performance in operational codebases.
The trend: Code-model competition is moving from general programming benchmarks toward workflow-specific performance on enterprise software that is difficult to modernize.