OpenAI Announces New AI Models for Low-Resource Languages

OpenAI has introduced new model capabilities supporting software development and modernization processes in low-resource programming languages with limited code and documentation, such as COBOL, Fortran, and Visual Basic. To measure performance in these languages, which are used in critical infrastructures globally, MultiLangBench and a COBOL migration assessment have been developed.

While COBOL powers over 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions, Fortran is used in weather forecasting and nuclear reactor safety analyses. The fact that these languages contain fewer resources compared to common languages like Python and JavaScript poses a significant challenge for artificial intelligence models.

Developed to measure progress, MultiLangBench tests whether models can build an SQLite-compatible database engine from scratch in 20 different programming languages. The COBOL migration assessment measures whether existing programs can be ported to another language while preserving their behaviors.

The new models exhibited high accuracy and low cost rates in both evaluations. While GPT-6 Astra achieved a test success rate of 87.4% on MultiLangBench, GPT-6 Luna provided a 40.3% success rate at a lower estimated API cost. In the COBOL migration test, GPT-6.1 Sol achieved an accuracy rate of 86.6% at low reasoning effort and 97.9% at high reasoning effort.