OpenAI Launches a Coding Model for Older Programming Languages

OpenAI announced GPT-6 Astra today. It is an AI model designed to write and work with code in COBOL, Fortran, Visual Basic, and other older programming languages.

“There is less code and written information available for models to learn from in these languages than in languages such as Python and JavaScript. However, these older languages still support important computer systems and essential services around the world,” OpenAI explains in its announcement. “Developers and large companies have told us that coding models often have trouble working with these languages. Their feedback helped guide this work.”

OpenAI created two tests, MultiLangBench and COBOL migration, to check whether its models can build new software in these older languages, which often have less written information available. The tests also check whether models can keep older apps working as they are updated or moved to newer programming languages. GPT-6 Astra scored 87.4 percent across 20 languages in MultiLangBench, compared with 26.8 percent for GPT-5.6 Sol. In the COBOL migration test, GPT-6 Astra scored 86.8 percent, rising to 97.9 percent when using high reasoning effort. GPT-5.6 Sol scored only 5.4 percent on that test.

“Our goal is to help models use their software engineering skills across more languages. This includes understanding code projects they have not seen before, planning changes, using tools to build software, running tests, and finding and fixing problems,” OpenAI says.

Developers can use GPT-6 Astra through the OpenAI API for coding tasks involving COBOL, Fortran, OCaml, and Visual Basic.