There is a wide disconnect between real world usage and leaderboards. If gemini was so good why are so few using them?

Having tested that model in many real world projects it has not once been the best. And going farther it gives atrocious nonsensical output.