TL;DR: OpenAI published new tests showing improved model performance in less-common programming languages and code migration tasks.
Some of the world’s most important software is written in languages that receive relatively little attention from modern coding tools. Banking infrastructure, scientific applications and older government systems still depend on COBOL, Fortran and Visual Basic. OpenAI’s October 19 research update focuses on making AI assistance more capable in those environments.
The company introduced its MultiLangBench evaluation and described substantial gains across 20 programming languages. In its published results, GPT-6 Astra achieved an average test pass rate of 87.4%, compared with 26.8% for GPT-5.6 Sol on the same benchmark. Separate tests examined how faithfully models can migrate COBOL code into modern languages.
The practical challenge is not merely translating syntax. Financial programmes may encode decades of assumptions about rounding, missing records and regulatory reporting. A migration that looks tidy could still introduce an error affecting real money if it changes those behaviours.
Benchmark results help identify progress, but they are not proof that an entire bank can automate modernisation without engineers, testing and independent review. More capable models could nevertheless make costly maintenance less intimidating, allowing institutions to inspect and gradually replace systems that have remained difficult to change for years.
Leave a Reply