OpenAI says its new models are better for building and migrating software in low-resource legacy programming languages like COBOL, Fortran, and Visual Basic
Context & Ripple Effects
OpenAI's legacy-language claim follows its 2025 SWE-Lancer benchmark, which found real-world freelance software work difficult for frontier models, and its later emphasis on coding benchmark results in GPT-5's reported evaluation scores. The shift matters because it moves the coding pitch from measured performance on broad tests toward building and migrating software where training material is limited.
Microsoft's June release of business-focused models, including a coding model fine-tuned for GitHub, shows that coding capability is being packaged around enterprise deployment rather than treated solely as a general-purpose model benchmark.
First-order effects
- OpenAI can position its models for organizations maintaining COBOL, Fortran, and Visual Basic software, extending its coding offering to both new development and migration work.
- Teams working in those languages gain a vendor-claimed AI option for tasks that are less represented in the model-training ecosystem.
Second-order effects
- Microsoft's business coding-model strategy faces a more specific competitive test: enterprise buyers can compare model support for legacy-language workflows, not only general coding performance.
- Migration projects become a more important proving ground for coding models, since buyers can assess whether a model helps translate and rebuild existing software rather than merely generate isolated code.
Third-order effects
- If legacy-language performance becomes repeatable, coding-model competition will shift from broad benchmark scores toward integration with the maintenance and modernization workflows that enterprises already run.
- The durable advantage may accrue to vendors that pair model capability with reliable migration processes, testing, and deployment systems rather than to those with coding scores alone.
The trend: AI coding tools are moving from general code generation toward workflow-native support for enterprise software modernization, including low-resource legacy stacks.