OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran
decoder ·
OpenAI has announced significant improvements in its latest GPT-6 models regarding their proficiency with legacy programming languages such as COBOL, Fortran, and Visual Basic. These older languages are critical for global infrastructure, with IBM estimating that COBOL supports the majority of ATM transactions and online banking systems. To measure this progress, OpenAI developed two evaluation benchmarks: MultiLangBench and a COBOL migration benchmark. On MultiLangBench, which requires building a database engine from scratch, GPT-6 Astra achieved an average pass rate of 87.4 percent across 20 languages, a substantial increase from GPT-5.6 Terra's 38 percent. The newer GPT-6 Luna model scored 40.3 percent but did so at a significantly lower cost compared to its predecessor. In the COBOL migration test, which involves translating code into modern languages like Python or Rust, GPT-6.1 Sol reached 86.6 percent accuracy with low reasoning effort. Increasing the reasoning effort for GPT-6.1 Sol further boosted its accuracy to 97.9 percent, demonstrating a dramatic leap over previous versions.
decoder