AI models make strides in supporting low-resource programming languages

Researchers have unveiled significant advancements in AI models capable of working with low-resource programming languages such as COBOL, Fortran, and Visual Basic. These languages, which support critical infrastructure like banking systems and weather forecasting, often lack extensive documentation and code for AI to learn from. The new models demonstrate improved accuracy and cost-efficiency in tasks like building software and migrating code while preserving functionality. Two evaluations, MultiLangBench and a COBOL migration test, were used to measure progress. GPT-6 Astra achieved an 87.4% pass rate on MultiLangBench, up from 26.8% for GPT-5.6 Sol. GPT-6 Luna also showed improvements, reaching 40.3% accuracy at a lower cost. On the COBOL migration test, GPT-6.1 Sol achieved 86.6% accuracy at low cost, with higher accuracy at increased reasoning effort. The results highlight the growing capability of AI to support a wide range of programming languages, offering developers tools to maintain and modernize essential software systems.