> ## Content Index
> Fetch the complete content index at: https://theainewsfiles.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# OpenAI Says Its Models Now Migrate COBOL Far More Accurately; Academics Tell The Verge the Hard Part Lies Elsewhere
- URL: https://theainewsfiles.com/openai-cobol-migration-claims-academics-sceptical/
- Published: 2026-10-26T15:35:07.000Z
- Updated: 2026-10-26T15:35:07.000Z
- Description: OpenAI says its latest models are stronger at migrating COBOL, and BleepingComputer reports OpenAI's figures showing far higher migration accuracy than GPT-5.6 Sol. Academics who spoke to The Verge say the test leaves out much of what makes replacing legacy systems hard.
- Author: J E Cowie
- Tags: News

On 19 October 2026, OpenAI said on X: "Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic." The post added that progress here "means understanding decades of business logic and helping developers modernize those systems while preserving what works." These are OpenAI's claims about its own models. This article separates what the company says it measured from what outside academics told The Verge those measurements leave out.

## What OpenAI says it measured

OpenAI's announcement covers several results. One of them is a test of COBOL migration. BleepingComputer reported on 20 October that OpenAI tested whether its models could migrate COBOL programs to C++, Java, Kotlin, Python or Rust without changing their behaviour.

OpenAI's figures, as quoted by 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." It added: "At higher reasoning effort, accuracy reaches 97.9%, bringing the evaluation close to saturation."

The Verge reports OpenAI acknowledging that "a migration is considered successful only if all tests pass." Tech Times describes the standard as passing hidden tests, including tests aimed at edge cases such as rounding and missing-record handling.

OpenAI also built a second test, MultiLangBench, which BleepingComputer says asks models to build a SQLite-compatible database engine from scratch across 20 programming languages. On it, BleepingComputer reports, GPT-6 Astra reached an average test pass rate of 87.4%, up from 26.8% for GPT-5.6 Sol. In COBOL specifically it moved from 10.9% to 72.1%.

![a cash machine (ATM) in use](https://theainewsfiles.com/content/images/2026/10/shotV2_img-2113.jpg)

a cash machine (ATM) in use

## Why COBOL matters

According to BleepingComputer, OpenAI says COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions and 95% of ATM transactions. It says Fortran is still used in areas such as weather forecasting and nuclear reactor safety analysis. The Verge attributes the same three COBOL figures to IBM.

OpenAI said in a blog post, quoted by BleepingComputer, that these languages have less code and documentation available for models to learn from than languages like Python and JavaScript. BleepingComputer quotes OpenAI as saying they nonetheless "support critical infrastructure and essential services around the world."

The Verge also reports that Anthropic published a "code modernization playbook" in February, pitching its Claude Code agent for COBOL migration. The Verge says it offered no performance benchmarks and declared that "the economics of COBOL modernization have shifted."

## What the academics say

The Verge, in a piece by Robert Hart published on 26 October, spoke to several university researchers. It says they agreed that AI tools have made strides on legacy code in recent years, but cautioned that OpenAI's results are a very long way from making it possible to safely replace critical legacy systems.

- **Vadim Zaytsev**, associate professor of software evolution at the University of Twente: "Rewriting code is not the same as replacing a system." He said the largest system he had seen had 250 million lines of COBOL, but that code is only part of the problem: systems can involve several languages, data in various formats, integrations with other systems and the expectations of the people and organizations involved. He said "AI is not playing a major role at the moment in replacing or migrating legacy COBOL systems."
- **Arie van Deursen**, software engineering professor at Delft University of Technology, called OpenAI's blog "superficial," saying it offered few details about the migration task. He called the tasks OpenAI used "lightyears" away from actual modernization, and said: "The evaluations shown by OpenAI don’t address the key difficulty: Breaking up a complex legacy system into smaller units that can be replaced individually."
- **Alastair Donaldson**, professor of programming languages at Imperial College London, said robust testing is crucial. The Verge describes one approach as running the same workflows through the old and new systems to check they behave identically.
- **Jacopo Mauro**, mathematics and computer science professor at the University of Southern Denmark, said formal verification of legacy code would require a "huge amount of work," and that until then "human verification remains essential."

Zaytsev also offered an image of the gap that remains: "Suppose you have a magic AI wand that handles 99 percent of your code." His question: "What do you do with the remaining 1 percent?" He said that 1 percent can be a million lines of code, spread all over the place.

The Verge reports that OpenAI spokesperson Oscar Haines acknowledged in an email that legacy code "is still far from a solved problem."

## What the test covers

Tech Times, in an article of 21 October, makes a similar point from the benchmark's side. It states that the COBOL migration evaluation tests single-file programs, and that copybooks, JCL job streams and undocumented business logic fall outside what it measures. BleepingComputer adds that the work could help companies modernise older software, although production migrations would still require expert review and testing.

My reading: the figures and the criticisms describe different things. OpenAI's numbers, as reported by BleepingComputer, describe how well a model migrates COBOL programs without changing their behaviour under a defined test. The academics are talking about replacing whole systems. The jump from 5.4% to 86.6% at the same low reasoning setting is OpenAI's own reported result. No source in this reporting says that result has yet been reproduced on real production systems.

## Sources

Hart, R. (2026) 'Can AI help solve the COBOL code crisis?', The Verge, 26 October 2026, available at: [https://www.theverge.com/ai-artificial-intelligence/1018434/cobol-migration-legacy-code-openai-anthropic](https://www.theverge.com/ai-artificial-intelligence/1018434/cobol-migration-legacy-code-openai-anthropic?ref=theainewsfiles.com) (Accessed: 26 October 2026).

Mitchell, J. (2026) 'GPT-6 Cracks COBOL: 97.9% Migration Accuracy Shifts AI Bottleneck from Code to Enterprise', Tech Times, 21 October 2026, available at: [https://www.techtimes.com/articles/329306/20261021/gpt-6-cracks-cobol-979-migration-accuracy-shifts-ai-bottleneck-code-enterprise.htm](https://www.techtimes.com/articles/329306/20261021/gpt-6-cracks-cobol-979-migration-accuracy-shifts-ai-bottleneck-code-enterprise.htm?ref=theainewsfiles.com) (Accessed: 26 October 2026).

OpenAI (2026) 'Our latest model advances bring stronger capabilities for building and migrating…', X post by @OpenAI, 19 October 2026, archived by Post-Cutoff, available at: [https://postcutoff.com/p/x-openai-2112310345863930168/](https://postcutoff.com/p/x-openai-2112310345863930168/?ref=theainewsfiles.com) (Accessed: 26 October 2026).

Parmar, M. (2026) 'OpenAI says GPT-6.1 Sol is dramatically better at migrating legacy code', BleepingComputer, 20 October 2026, available at: [https://www.bleepingcomputer.com/news/artificial-intelligence/openai-says-gpt-61-sol-is-dramatically-better-at-migrating-legacy-code/](https://www.bleepingcomputer.com/news/artificial-intelligence/openai-says-gpt-61-sol-is-dramatically-better-at-migrating-legacy-code/?ref=theainewsfiles.com) (Accessed: 26 October 2026).

## Picture credits

- Top picture. Photo: Jim Evans, [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/?ref=theainewsfiles.com), via [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:National%5FBiscuit%5FCompany%5FBuilding%5F%28Nabisco%29%5F--%5FHouston.jpg?ref=theainewsfiles.com)
- Photo: Wil540 art, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:A%5Fperson%5Fuses%5Fan%5FATM%5Fat%5FChase%5FBank,%5FNYC%5F-%5F2008.tif?ref=theainewsfiles.com) ([CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/?ref=theainewsfiles.com))