Can AI solve the COBOL legacy code problem? Experts say not yet
OpenAI reports gains in migrating old code, but researchers say replacing critical systems is still far off.
Code#04
98%
GPT-6.1 Sol accuracy at higher reasoning effort
Image: The Verge
98%GPT-6.1 Sol accuracy at higher reasoning effort
The short version
- OpenAI said its latest models improved at migrating aging software from older languages like COBOL into modern code.
- In OpenAI's tests, GPT-6.1 Sol reached 87 percent behavior preservation accuracy at low reasoning effort and 98 percent at higher reasoning.
- Anthropic earlier pitched its Claude Code agent for COBOL migration but did not publish performance benchmarks.
- Experts told The Verge that the tests do not capture the difficulty of breaking up complex legacy systems.
Why it matters
COBOL still runs parts of banking, card payments, and government systems, so migration tools matter for developers. The experts quoted say understanding the system and verifying behavior are the main obstacles, not rewriting code alone.
“Rewriting code is not the same as replacing a system”