OpenAI's New Model Masters Niche Programming Languages, COBOL Migration Accuracy Soars to 97.9%
OpenAI's latest models improve development and migration for low-resource programming languages such as COBOL, Fortran, and Visual Basic. In a SQLite engine build test across 20 languages, GPT-6 Astra's average pass rate rose from 26.8% (GPT-5.6 Sol) to 87.4%; GPT-6 Luna reached 40.3%, with API cost down 78%. In COBOL migration tests, accuracy hit 97.9% at high reasoning effort.
OpenAI's New Model Tackles Underrepresented Programming Languages, COBOL Migration Accuracy Soars to 97.9%
OpenAI's latest model substantially improves development and migration capabilities for low-resource programming languages such as COBOL, Fortran, and Visual Basic. In a 20-language test involving building a SQLite engine from scratch, GPT-6 Astra's average pass rate jumped from 26.8% for GPT-5.6 Sol to 87.4%; the low-cost GPT-6 Luna also reached a 40.3% pass rate, with API costs plunging 78%.
In tests migrating legacy COBOL code to modern languages, GPT-6.1 Sol's accuracy surged from 5.4% to 86.6% even at low reasoning effort, with costs reduced by 39%; at high reasoning effort, accuracy reached as high as 97.9%. The relevant evaluation set is now open-sourced on GitHub.
OpenAI: https://openai.com/index/low-resource-programming-languages/
