OpenAI released GPT-6 models on October 19, 2026, saying they significantly improve the ability to work with low-resource programming languages such as COBOL, Fortran, and Visual Basic. It is the company's first major model update since the release of GPT-5.6 Sol in 2025.
OpenAI reported an average test pass rate of 87.4% on MultiLangBench, measured across 20 programming languages. That compares with 26.8% for GPT-5.6 Sol. GPT-6 Luna also achieved 40.3% accuracy at 78% lower estimated API cost.
"Building SQLite from scratch across 20 languages is a strong indicator of a model's ability to understand and implement complex software systems," said OpenAI. The evaluation measures how much working functionality a model can build over an extended coding session.
The announcement follows OpenAI's focus on improving AI's capabilities for enterprise software, as highlighted in direct feedback from developers and enterprise customers. OpenAI said the progress demonstrates a move toward more accurate and affordable code migration for teams modernizing critical systems.
OpenAI did not say when the models will be available for all users, and noted the need for expert review and testing when applying these capabilities to production software. The company plans to release the MultiLangBench evaluation setup and test suite on GitHub.
Source: openai
