That's weird. I was complaining before that GPT-5.1-Pro wasn't as good as Gemini 3 Pro, but today a few tasks made me feel like GPT-5.1-Pro has clearly improved, while Gemini 3 Pro seems to have regressed. Did they secretly switch models or tweak parameters behind the scenes?