OpenAI’s GPT-6 Astra boosts COBOL migration for banks—but expert oversight is critical

OpenAI’s AI models target legacy code migration

OpenAI has announced that its latest AI models, including GPT-6 Astra, now offer improved capabilities for translating and migrating software from legacy programming languages such as COBOL, Fortran, and Visual Basic. According to Engadget, this development targets institutions—particularly banks—that still rely heavily on these outdated systems but face a shrinking pool of skilled programmers.

COBOL’s persistent role in global banking

Engadget reports that 40% of the world’s banks continue to depend on COBOL, a language developed in the 1950s. The cost of migrating these systems to modern languages like Python or Java is estimated to range from millions to hundreds of millions of dollars per institution. Adding to the urgency, COBOL programmers are retiring, leaving fewer experts available to maintain or update these critical systems.

MultiLangBench tests reveal AI strengths and limits

To evaluate its models’ capabilities, OpenAI created MultiLangBench, a test suite designed to assess AI performance in generating SQLite-compatible database engines across 20 programming languages. According to Engadget, GPT-6 Astra achieved an average pass rate of 87.4% in these tests. However, COBOL remained the most challenging language, with a pass rate of 72%.

High accuracy in COBOL migration, but caveats remain

OpenAI also tested its models’ ability to migrate COBOL programs to languages like C++, Java, Kotlin, Python, and Rust while preserving their original functionality. Engadget notes that GPT-6.1 Sol, the latest iteration, achieved a 97.9% accuracy rate in these migrations when operating at high reasoning effort. At lower reasoning effort, the accuracy dropped to 86.6%.

Expert oversight still critical for AI migration

Despite these advancements, OpenAI emphasizes that AI-driven migration is not a substitute for expert review. The company warns that even small errors in migrated banking software could have severe consequences, necessitating thorough testing before deployment in production environments.

Why COBOL remains entrenched in enterprise systems

COBOL (Common Business-Oriented Language) has been a cornerstone of enterprise computing since the 1960s, particularly in industries like banking, finance, and government. Its resilience stems from its reliability and the high cost of replacing systems that handle critical operations, such as transaction processing and payroll.

The evolution of legacy system modernization

The challenge of modernizing legacy systems is not new, but the retirement of experienced COBOL programmers has accelerated the need for solutions. Previous attempts at automated migration have often relied on rule-based tools, which struggle with the complexity and nuances of decades-old codebases. AI-driven approaches, like those introduced by OpenAI, aim to address these limitations by leveraging large language models to understand and translate code more accurately.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

For founders and operators in banking and enterprise software, OpenAI’s advancements in AI-driven code migration are a double-edged sword. On one hand, they offer a potential lifeline for legacy systems that are becoming increasingly unmanageable as COBOL experts retire. The promise of reducing migration costs—even if just marginally—could be a game-changer for institutions facing budget constraints.

However, the 72% pass rate for COBOL, while impressive for an AI model, underscores that these tools are not a silver bullet. The stakes in financial systems are too high to rely solely on automated migration, no matter how advanced. OpenAI’s emphasis on expert review and testing is a necessary caveat, but it also means that banks will still need to invest in skilled engineers to validate and refine the output. For founders building tools in this space, the opportunity lies in hybrid solutions that combine AI efficiency with human oversight, rather than promising fully autonomous migration.

The broader takeaway? AI is accelerating the modernization of legacy systems, but the transition will still require careful planning, significant resources, and a healthy dose of skepticism about what the technology can actually deliver.

Key takeaways

• Banks and enterprises with legacy systems must prioritize migration planning, as COBOL expertise is disappearing and costs will rise over time.
• AI-driven migration tools like OpenAI’s GPT-6 Astra can reduce effort and cost but are not foolproof—expert validation remains critical.
• Founders building in this space should focus on hybrid solutions that pair AI efficiency with human oversight to mitigate risks.
• The high stakes of financial systems mean even a 97.9% accuracy rate may not be sufficient for full automation without rigorous testing.
• This development signals a growing market for legacy modernization tools, but success will depend on transparency about limitations.

FAQ

Why are banks still using COBOL?

COBOL remains widely used in banking because it is highly reliable for critical operations like transaction processing and payroll. Replacing these systems is expensive and risky, so many institutions have opted to maintain them instead.

How accurate is OpenAI’s AI in migrating COBOL code?

According to OpenAI’s tests, GPT-6 Astra achieved a 72% pass rate for COBOL migration, while GPT-6.1 Sol reached 97.9% accuracy at high reasoning effort. However, OpenAI stresses that these results require expert validation before deployment.

What is MultiLangBench?

MultiLangBench is an AI evaluation suite created by OpenAI to test models’ ability to generate and migrate SQLite-compatible database engines across 20 programming languages, including COBOL and Fortran.

Why is AI-driven migration not enough on its own?

Even small errors in migrated code can have severe consequences, particularly in financial systems. OpenAI recommends thorough testing and expert review to ensure the integrity of the migrated software.

Related on Lazyfounder

• OpenAI
• More technology news
• Arm’s AGI CPU Outperforms AMD in Agentic AI Benchmarks, Signal65 Reports
• OpenAI’s mathematical breakthroughs, security failures, and legal troubles raise industry alarms