Advancing coding in low‑resource programming languages
What happened
OpenAI evaluated its new GPT‑6 models on two new benchmarks that focus on low‑resource programming languages such as COBOL, OCaml, Fortran and Visual Basic. The MultiLangBench test asks a model to build a SQLite‑compatible engine from scratch in 20 languages, while a separate COBOL migration test measures how well a model can translate COBOL code to modern languages while preserving behavior. GPT‑6 Astra achieved an average 87.4% pass rate on MultiLangBench, up from 26.8% for the prior GPT‑5.6 Sol, and showed large gains across all languages. GPT‑6 Luna offered a lower‑cost option with a 40.3% pass rate at about 78% lower API cost. On the COBOL migration task, GPT‑6.1 Sol reached 86.6% accuracy at low reasoning effort, far surpassing the 5.4% of GPT‑5.6 Sol, and could hit 97.9% with higher effort.
Why it matters
These improvements mean developers maintaining legacy systems can rely more on AI to modernize code safely and cheaply. Businesses that depend on old software may see reduced risk and lower maintenance costs.
Reality check
An automated, sceptical read of the sourceIncremental. The results show substantial performance and cost gains, but they extend existing capabilities rather than introducing a fundamentally new technology.
Is this a big deal?
Get stories like this every morning
The day's AI news in plain English, in your inbox at 8:30 IST. Free.
More from OpenAI
Introducing ChatGPT WalletOpenAI · 13 Nov 2026Shield helps technicians hand off legacy portal work with OpenAIOpenAI · 13 Nov 2026OpenAI launches ChatGPT Wallet, AI agents pay for purchaseOpenAI · 13 Nov 2026OpenAI disbands its robotics team as its general models start running robotsOpenAI · 13 Nov 2026Yang-Mills, OpenAI proof, and the mass gap problemOpenAI · 13 Nov 2026This page was collected from openai.com and summarised automatically by AIPulsee's pipeline. See how it runs.