Thesis
OpenAI’s announcement is evidence of a targeted capability push for software work in low-resource programming languages, but not proof of reliable legacy-system migration in production.
Observed facts
OpenAI says it has improved ChatGPT for building and migrating software in COBOL, Fortran, and Visual Basic.
1The announcement describes new evaluations covering software engineering, SQLite, and COBOL migration.
1The source says these languages underpin software used by banks, governments, and businesses.
1
Mechanism
The update pairs model capability work with evaluations focused on language- and task-specific software engineering, making progress potentially more legible through defined tasks rather than broad coding claims.
Why now
OpenAI’s announcement puts languages associated with long-lived institutional software inside its current coding capability agenda, while naming both build and migration work.
Implications
The first implication is that teams should validate any claimed assistance against their own codebases and migration acceptance criteria.
A benchmark or model improvement alone does not establish correctness across a production system.
Evidence boundary
The supplied primary source states the language scope and names evaluation areas, but the provided material gives no quantitative results, independent replication, customer deployment evidence, or proof of end-to-end migration reliability.
Watch conditions
The first watch condition is whether OpenAI publishes evaluation methods and results that show performance on representative, reviewable migration tasks.
Whether independent or customer evidence reports outcomes on real legacy code.
Whether workflows retain human verification for consequential changes.