OpenAI says its models are better at migrating COBOL systems
The company is using that claim to target banks still dependent on old code, where modernization has been slow and risky.
OpenAI says its newer models are much better at handling legacy languages, and it is pitching that improvement to banks and other enterprises that still depend on COBOL-based systems. The company says COBOL remains embedded in a large share of banking infrastructure, including more than 40% of online banking systems and 95% of ATM transactions, which makes migration both costly and urgent as COBOL specialists age out. To measure the jump, OpenAI built MultiLangBench, a test that asks models to create SQLite-compatible database engines in 20 languages, and GPT-6 Astra scored 87.4% on average versus 26.8% for GPT-5.6 Sol. On COBOL specifically, Astra rose to 72.1%, while GPT-6.1 Sol reached 86.6% accuracy at low reasoning effort and 97.9% at higher effort when migrating COBOL code to C++, Java, Kotlin, Python or Rust. OpenAI says these results could help modernize older software, but any production migration still needs expert review and testing against business rules and dependencies.
Why it matters
For banks, the practical shift is that more of the first pass at legacy-code migration can be automated, rather than starting from scratch with specialist teams. That does not remove the need for review, but it can reduce the amount of manual work needed to assess and translate older systems that still sit inside core banking infrastructure.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.