OpenAI Models Promise to Modernize Legacy Software Languages

OpenAI announced that its latest artificial intelligence models, including GPT-6 Astra and GPT-5.6 Sol, have shown significant improvements in translating legacy programming languages like COBOL, used in critical infrastructure such as banking and public services, into newer languages.

In new research guided by developer and enterprise customer feedback, OpenAI created two new benchmarks to test its models' ability to transform legacy systems. The first test, MultiLangBench, measures how successfully models can create a SQLite program from scratch in a given coding language, with GPT-6 scoring the highest with an average success rate of approximately 87 percent. The second test evaluated whether models could migrate COBOL-based programs into languages like C++, Java, Kotlin, Python, or Rust without meaningfully changing the program's behavior. According to OpenAI, GPT-6.1 Sol with high reasoning effort achieved a score extremely close to 100 percent in this test.