OpenAI GPT 6 1 COBOL Migration

A COBOL migration benchmark went from 5.4% to 97.9% with OpenAI's GPT-6.1 Sol.

That is the reported score at maximum reasoning effort, and the caveat matters. But the size of the jump is what stands out.

Legacy code is one of the most expensive problems in enterprise tech. Banks, insurers and government agencies run critical systems on COBOL that few engineers still want to maintain. For years the migration business case failed on one question: could anyone trust the output?

A score near 98% changes that question. The constraint moves from "can the model translate this" to "can we verify, test and govern what it produces."

Here is my read for leaders:

1. Migration timelines built on old assumptions are now worth re-pricing. 2. The bottleneck shifts to test coverage, validation and domain knowledge of what the old system was supposed to do. 3. Benchmarks at maximum reasoning effort are a ceiling, not a production guarantee. Pilot on your own code before you plan around the number.

The organizations that benefit most will not be the ones with the best model access. They will be the ones with the best documentation and test suites around their legacy systems.

If a 97.9% score held up on your own codebase, what would you modernize first?