Censorship is a given. What's the actual performance on benchmarks? Such as for code generation. Or can we not agree on benchmarks because of the language diff

Censorship is a given. What's the actual performance on benchmarks? Such as for code generation. Or can we not agree on benchmarks because of the language difference?