capabilitiesIn brief · · · Written with AI
OpenAI says its models are far better at coding in COBOL and Fortran
The company published two of its own tests on languages with little public code. Its newer models go from 5% to 87% on migrating COBOL programs.
On 15 October, OpenAI published the results of two of its own tests on programming in languages with little public code, such as COBOL, Fortran and Visual Basic. The company says these languages keep critical services running: COBOL, for example, supports more than 40% of online banking systems.
On MultiLangBench, which asks a model to build a SQLite-compatible database engine in 20 languages, GPT-6 Astra passes 87.4% of tests on average. GPT-5.6 Sol passed 26.8%. On a test of migrating COBOL programs to other languages, GPT-6.1 Sol reaches 86.6% at low reasoning effort. GPT-5.6 Sol scored 5.4%.
These are the company's own figures, on evaluations it designed itself. OpenAI acknowledges that applying such migrations to real software still takes expert review and testing.
Sources
Written by AI from these sources.
- Advancing coding in low-resource programming languagesAccessed 20 October 2026 · Image: OpenAI
