It _is_ better in the general case on most benchmarks. There are also very likely specific use cases for which it is worse and very likely that OpenAI doesn't k

It _is_ better in the general case on most benchmarks. There are also very likely specific use cases for which it is worse and very likely that OpenAI doesn't know what all of those are yet.