The metrics being the benchmarks? Perhaps but apparently these are being gamed. Which makes sense. It's the natural tendency of people.

But in my daily usage? It's a fully different story.

They don’t. GPT 5.2 and its variants are the best models right now.