Artificial intelligence is tackling a massive translation task that could transform critical infrastructure in banking, nuclear energy, and government agencies. OpenAI announced that its latest models, including GPT-6 Astra and GPT-5.6 Sol, have demonstrated substantial improvements in migrating old programming languages into newer ones. This capability addresses a critical vulnerability, as many foundational services still rely on dated codebases.
COBOL, a language invented in 1959, still handles 95 percent of all ATM transactions, while Fortran, created in 1957, is widely used to simulate nuclear reactor functionality. However, engineers trained in these legacy languages are retiring, creating a widening gap as younger programmers utilize AI tools primarily trained on contemporary software. To address this, OpenAI developed new evaluation benchmarks, including MultiLangBench, to test how well models can build programs like SQLite from scratch using specific coding languages.
The company also tested whether models can successfully migrate COBOL-run programs into newer languages such as C++, Java, Kotlin, Python, or Rust without altering behavior. According to OpenAI, its reasoning models scored near 100 percent on tests requiring the preservation of strict rounding rules and broader contextual logic. Competitors are also pursuing this space; Anthropic previously highlighted similar capabilities for Claude Code, which it noted could help businesses modernize legacy systems in quarters rather than years.
Comments