OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran

The ability to reliably translate legacy code into modern languages lowers the risk and cost of maintaining critical systems.

OpenAI announced advancements in its GPT-6 models, particularly in handling legacy programming languages like COBOL and Fortran. These languages are crucial for maintaining critical global infrastructure, with COBOL powering over 40 percent of online banking and 95 percent of ATM transactions. To measure progress, OpenAI developed MultiLangBench, which tests database engine creation. GPT-6 Astra achieved an 87.4 percent pass rate across 20 languages, a significant jump from GPT-5.6 Terra's 38 percent. A variant, GPT-6 Luna, performed comparably on this benchmark but at a lower cost. Another benchmark focused on migrating COBOL code to modern languages. GPT-6.1 Sol demonstrated 97.9 percent accuracy in COBOL to C++, Java, Kotlin, Python, or Rust translation with high reasoning effort, vastly outperforming GPT-5.6 Sol's 5.4 percent accuracy.

These performance gains in legacy code processing signal a maturing AI capability that directly addresses a significant technical debt burden across global industries. The ability to accurately migrate COBOL and Fortran code means that companies relying on these foundational, yet aging, systems can modernize more safely and affordably. This development could accelerate digital transformation projects that were previously stalled due to the prohibitive cost and risk associated with manual or less capable AI-assisted legacy code migration.

The success of GPT-6 in COBOL and Fortran transformations suggests a broader trend towards AI models becoming indispensable tools for infrastructure maintenance and modernization, not just for new development. This advancement will likely spur increased investment in AI solutions for enterprise software and potentially create new opportunities for AI-powered managed services focused on system upgrades and security patching for older systems. Companies that have been hesitant to adopt AI due to concerns about its applicability to their core legacy systems may now reconsider.