OpenAI's GPT-6 Astra posts 87.4% pass rate on legacy-code tests
The model targets COBOL, Fortran and Visual Basic, and OpenAI says it averages an 87.4% pass rate on its new 20-language benchmark, MultiLangBench.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
What happened
OpenAI announced GPT-6 Astra, an AI coding model tailored to COBOL, Fortran, Visual Basic and other legacy languages. On MultiLangBench, a new benchmark covering 20 languages, the model reached an 87.4 percent pass rate across its tests. That compares with about 38 percent for the earlier GPT-5.6 Terra.
Why it matters
The figures indicate a large performance gain over the prior model on older programming languages.
Coverage
- OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran (the-decoder.com)
- OpenAI Launches a Coding Model for Older Programming Languages (thurrott.com)
- OpenAI reports 87.4% MultiLangBench pass rate for GPT-6 Astra (developer-tech.com)
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations