OpenAI says GPT-6.1 Sol scores 86.6% in translating COBOL

New model scores 86.6% in translating COBOL to modern languages, up from 5.4%

By LineZotpaper

Published20 October 2026

Updated20 October 2026

Read Time2 min

Sources2 outlets

OpenAI's latest large language model, GPT-6.1 Sol, represents a significant leap in the ability to work with legacy programming languages. In benchmarks measuring the migration of COBOL programs to modern languages like C++, Java, and Python, the new model achieved 86.6% accuracy at a low reasoning effort, a dramatic improvement over its predecessor's 5.4%, the company has stated.

OpenAI confirmed that its newer models are much better at writing and migrating software in older programming languages. The company lists programming languages such as COBOL, Fortran, and Visual Basic as a few examples where it has seen significant improvements.

According to OpenAI, COBOL supports over 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions, while Fortran is still used in areas such as weather forecasting and nuclear reactor safety analysis. These languages support critical infrastructure around the world despite having less training data available than more modern languages.

To measure progress, OpenAI created MultiLangBench, which asks models to build a SQLite-compatible database engine from scratch across 20 programming languages. GPT-6 Astra reached an average test pass rate of 87.4%, up from 26.8% for the previous model. Improvements extended to every language tested, including 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%).

OpenAI also tested whether the model could migrate COBOL programs to C++, Java, Kotlin, Python, or Rust without changing their behaviour. GPT-6.1 Sol achieved 86.6% accuracy at low reasoning effort, compared with GPT-5.6 Sol's 5.4% at the same setting, at just 61% of the API cost. At higher reasoning effort, accuracy reached 97.9%, bringing the evaluation close to saturation. The company noted that production migrations would still require expert review and testing.

Analysis

Why This Matters

A massive body of critical financial and infrastructure software is written in languages like COBOL and Fortran, where expert programmers are scarce and expensive.
Practical AI-driven migration could unlock modernization without fully rewriting systems, improving security and efficiency.
The sharp drop in cost and rise in accuracy makes automated legacy migration a commercially viable prospect for the first time.

Background

Legacy code written in COBOL and Fortran runs a significant portion of the world's core financial systems, from bank ledgers to ATM networks. As skilled programmers in these languages retire, the need to migrate to modern, maintainable platforms has grown, but automated tools have historically struggled with the limited training data available for these older languages.

Key Perspectives

OpenAI: Argues the results demonstrate a new capability to modernize critical systems, significantly reducing the cost and risk of migration projects.

Financial Institutions and Infrastructure Operators: Face an opportunity to tackle deeply entrenched technical debt, potentially unlocking faster feature development and improved security postures.

Critics and Engineers: Warn that even a 97.9% accuracy rate still leaves room for subtle bugs, which in safety-critical or financial systems could be catastrophic. Human oversight and robust testing will remain mandatory.

What to Watch

Adoption trends among major banks and government agencies that rely on COBOL-based mainframes.
The development of specialized review and testing workflows built around AI-generated legacy migrations.
Any reported incidents of bugs introduced by AI migration in production environments.

Sources

OpenAI says GPT-6.1 Sol is dramatically better at migrating legacy code · BleepingComputer
OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran · The Decoder

Checked

12 claims in this story checked against 1 source: 10 supported.

Show each claim

SupportedOpenAI says GPT-6.1 Sol scores 86.6% accuracy at low reasoning effort on COBOL migration.BleepingComputer: ““GPT-6.1 Sol achieves 86.6% accuracy at low reasoning effort, compared with GPT-5.6 Sol's 5.4% at the same setting at just 61% of the API cost,” OpenAI said.”
SupportedThe model scores 86.6% translating COBOL to modern languages, up from 5.4%.BleepingComputer: ““GPT-6.1 Sol achieves 86.6% accuracy at low reasoning effort, compared with GPT-5.6 Sol's 5.4% at the same setting at just 61% of the API cost,” OpenAI said.”
Not in the sourcesGPT-6.1 Sol is OpenAI's latest large language model.
SupportedGPT-6.1 Sol's API cost is 61% of the comparison model's cost.BleepingComputer: ““GPT-6.1 Sol achieves 86.6% accuracy at low reasoning effort, compared with GPT-5.6 Sol's 5.4% at the same setting at just 61% of the API cost,” OpenAI said.”
SupportedAt higher reasoning effort, GPT-6.1 Sol's accuracy reaches 97.9%.BleepingComputer: ““At higher reasoning effort, accuracy reaches 97.9%, bringing the evaluation close to saturation.””
SupportedMultiLangBench asks models to build a SQLite-compatible database engine from scratch across 20 programming languages.BleepingComputer: “To measure progress, OpenAI created MultiLangBench, which asks models to build a SQLite-compatible database engine from scratch across 20 programming languages.”
SupportedGPT-6 Astra's average test pass rate was 87.4%, up from 26.8% for the previous model.BleepingComputer: “GPT-6 Astra reached an average test pass rate of 87.4%, up from 26.8% for GPT-5.6 Sol.”
SupportedCOBOL's test result improved from 10.9% to 72.1%.BleepingComputer: “In COBOL specifically, Astra jumped from 10.9% to 72.1%”
SupportedCOBOL supports over 40% of online banking systems.BleepingComputer: “OpenAI says COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions”
SupportedCOBOL supports 80% of in-person credit card transactions.BleepingComputer: “OpenAI says COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions”
SupportedCOBOL supports 95% of ATM transactions.BleepingComputer: “OpenAI says COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions, and 95% of ATM transactions”
Partly supportedThese languages have less training data available than more modern languages.BleepingComputer: ““These languages have less code and documentation available for models to learn from than languages like Python and JavaScript, yet support critical infrastructure and essential services around the world,” OpenAI explained in a blog post.”