On 19 October 2026, OpenAI said its latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, Visual Basic, and OCaml. These languages typically have less code and documentation available for models to learn from than languages like Python and JavaScript, yet they underpin critical infrastructure and essential services. OpenAI cites that COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions, while Fortran is used in weather forecasting and nuclear reactor safety analysis. The work is also guided by feedback from developers and enterprise customers that coding models often struggle with these languages—incomplete documentation, departed original authors, and business rules that exist only in code all make maintenance and modernization hard.
Two evaluations: building from scratch, and behavior-preserving migration
To measure progress, OpenAI developed two complementary evaluations. MultiLangBench asks a model to build a SQLite-compatible database engine from scratch in a specified language: organizing a codebase, managing state, handling errors, and adding features while preserving existing functionality. To compare across languages, the evaluation uses database files as a shared interface—the candidate receives an initial database and a SQL query, then produces a modified database checked against a reference SQLite implementation. Tests were adapted from SQLite’s existing suite; the evaluation reports both test pass rate and API cost.
The second evaluation asks models to migrate COBOL programs to C++, Java, Kotlin, Python, or Rust while preserving what each program does. Successful migration requires understanding both business logic and COBOL-specific behavior—for example, COBOL’s fixed-point arithmetic discards excess decimal digits unless rounding is explicitly requested; a rewrite that rounds instead can run without errors while producing different financial results. Hidden tests require byte-for-byte parity on observable outputs; a migration succeeds only if all tests pass. OpenAI notes that behavior-preserving capabilities like these are useful for teams modernizing older systems, though applying them to production software still requires expert review and testing against the system’s business rules and dependencies.
MultiLangBench: Astra averages 87.4%, above 83% in 19 of 20 languages
At maximum reasoning effort, GPT‑6 Astra achieves an average test pass rate of 87.4% across 20 languages, up from 26.8% for GPT‑5.6 Sol, with improvements in every language tested. Examples include COBOL (10.9% to 72.1%), OCaml (25.6% to 85.8%), Fortran (21.4% to 83.5%), and Visual Basic (27.5% to 89.4%). Astra scores above 83% in 19 of the 20 languages, with COBOL remaining the hardest at about 72%.
At the same maximum-effort setting, lower-cost GPT‑6 Luna reaches an average of 40.3% and also beats GPT‑5.6 Sol in every language tested, at an estimated average API cost of about $0.77 per sample versus about $3.48 for Sol—approximately 78% lower. OpenAI is releasing the MultiLangBench evaluation setup and test suite on GitHub, along with per-language results, so developers can inspect what the evaluation tests and where gaps matter for their work.
Figure: Left — colored language streams converging onto a shared bridge as a migration metaphor; right — a shared database-file interface with multiple soft “engines” connected in a build-lab scene — editorial illustration with no text or logos; not a screenshot of a real system.
COBOL migration: 86.6% at low effort, nearly 98% at higher effort
On the COBOL migration evaluation, GPT‑6.1 Sol achieves 86.6% accuracy at low reasoning effort, compared with 5.4% for GPT‑5.6 Sol at the same setting, while costing about 39% less (the detailed section also states about 61% of the prior API cost—equivalent figures). At higher reasoning effort, accuracy reaches 97.9%, nearly saturating the evaluation. OpenAI concludes that its models are becoming significantly better across a wider range of programming languages while also becoming more cost-efficient, and invites teams to try the latest models via the API on the languages they use and to share feedback on where focus should come next.
Primary source:
- OpenAI, “Advancing coding in low‑resource programming languages”, 2026-10-19, https://openai.com/index/low-resource-programming-languages/
