They are not equal or better. Try them out on real things. I did - it's not close. The benchmarks are a misrepresentation.

That being said- hard to believe OpenAI etc couldn't build a frontier classifier too.