What the author reports
Boris Cherny reports nearly complete HAProxy test-suite passes from both models. In the reported setup, Opus 5.5 completed the port in 9.5 hours versus 12 hours for Fable 5.1, at 51% lower cost. This is an Anthropic employee’s selected experiment, not our benchmark.
Supporting documentation (opens in a new tab)Why it belongs in the comparison
A long migration tests more than code fluency: keeping track of a system, responding to failures and reaching an inspectable stopping point. For your own migration, compare accepted changes and unresolved failures as well as runtime and spend.
What remains unknown
We have not reviewed the generated patches, exact settings, test coverage or production behavior. Passing nearly all existing tests is not proof of security or complete compatibility. Treat this as a lead for a bounded experiment, not a reason to replace a production component unchecked.