OpenAI's GPT-6.1 Sol Dramatically Boosts Legacy Code Migration
Original source
OpenAI says GPT-6.1 Sol is dramatically better at migrating legacy code
BleepingComputer →OpenAI’s latest models have made substantial progress in handling legacy programming languages like COBOL, Fortran, and Visual Basic—languages that remain fundamental to critical infrastructure worldwide. COBOL alone powers over 40% of banking systems, manages 80% of credit card transactions, and runs 95% of ATM machines. Despite their ubiquity, these older languages present a training challenge for AI models: limited code and documentation compared to modern languages like Python and JavaScript.
To quantify its advances, OpenAI built MultiLangBench, a test requiring models to implement a SQLite-compatible database engine across 20 programming languages. GPT-6 Astra achieved an 87.4% pass rate, a dramatic jump from GPT-5.6 Sol’s 26.8%. The improvements were consistent across all tested languages, with COBOL rising from 10.9% to 72.1%, Fortran from 21.4% to 83.5%, and Visual Basic from 27.5% to 89.4%.
More practically, GPT-6.1 Sol demonstrated strong capability at translating COBOL programs to modern languages, reaching 86.6% accuracy at low reasoning effort—compared to 5.4% for its predecessor—and 97.9% at higher reasoning effort. While these results suggest a viable path for modernizing legacy systems, OpenAI acknowledges that production migrations would still require specialist review and validation.
Read the full article
Continue reading at BleepingComputer →This is an AI-generated summary. Read the original for the full story.