OpenAI recently announced that its newer models are significantly better at writing and moving software built on older programming languages. The company revealed that GPT-6.1 Sol shows major improvements when working with legacy code like COBOL and Fortran. This marks a notable step forward for systems that support critical infrastructure and financial services around the world where older languages still run behind the scenes.
GPT-6.1 Sol boosts accuracy when migrating older programming languages
To test these new capabilities, the artificial intelligence company looked at how well its models could translate COBOL programs into modern languages like Python, Java, C++, Kotlin, or Rust without changing how the software behaves. The results showed massive gains.
At a low reasoning effort setting, GPT-6.1 Sol hit an 86.6 percent accuracy rate. For comparison, the older GPT-5.6 Sol model only managed a 5.4 percent accuracy rate under the same conditions.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
Not only is the newer model more capable, but it also achieves these results at just 61 percent of the previous API cost. If developers push the model to a higher reasoning effort, accuracy climbs to 97.9 percent.
Testing highlights major improvements across multiple older coding languages
Older programming languages often lack the vast amounts of code and documentation that models rely on to learn. Yet, languages like COBOL still support over 40 percent of online banking systems and 95 percent of ATM transactions. Fortran remains essential for weather forecasting and nuclear safety analysis.
To measure progress across the board, the maker of ChatGPT created a new benchmark called MultiLangBench. This test asks AI models to build a database engine from scratch using 20 different programming languages. On this benchmark, GPT-6 Astra achieved an 87.4 percent average pass rate, a massive jump from the 26.8 percent pass rate seen with GPT-5.6 Sol.
While these new models make updating old software much easier, companies will still need human experts to review and test the code before it goes live to prevent costly errors.
Discussion