Home / Technology / AI Tackles COBOL: Banks' Aging Code Gets a New Life
AI Tackles COBOL: Banks' Aging Code Gets a New Life
20 Oct
Summary
- AI models now proficient in COBOL and other legacy programming languages.
- 40% of global banks rely on COBOL, facing costly and time-consuming migrations.
- New AI achieves 87.4% pass rate in language tests, significantly improving performance.
AI models are demonstrating significantly improved capabilities in handling and migrating software written in older programming languages, such as COBOL, Fortran, and Visual Basic. This is particularly impactful for the financial sector, where approximately 40% of the world's banks still operate on COBOL systems. Migrating these systems to modern languages presents a substantial challenge, with projected costs reaching hundreds of millions of dollars.
OpenAI has developed advanced AI evaluations, including MultiLangBench, to test these capabilities. Their latest model, GPT‑6 Astra, achieved an impressive 87.4% average pass rate across 20 languages, a stark improvement over previous versions. While COBOL remains a difficult language, Astra still managed a 72% pass rate.
Furthermore, the AI has shown proficiency in migrating COBOL programs to C++, Java, Kotlin, Python, or Rust, crucially preserving the original program behaviors. GPT-6.1 Sol achieved 97.9% accuracy in preserving these behaviors, a vital factor for maintaining accuracy in banking operations. OpenAI notes that while these AI advancements are promising for modernizing older systems, production deployment still requires expert review and thorough testing.
