Wtf, Grok 4.20 is extremely high effort. Here's an example: One thing I love about Gemini is that if it recognizes concepts or terms, it will name them. "Hey it sounds like you're talking about Baumol's Cost Disease" without any tinkering with the system prompt. Claude and ChatGPT are conceited about this kinda thing. You basically have to BEG Claude not to treat you like a toddler or use a bunch of style guides. ChatGPT is the same. Grok is just like "oh yeah, here's the named concept, I went ahead and looked up the math, and did a bunch of number crunching, and here's the conclusion" Grok 4.20 default feels a lot like using ChatGPT Pro. No, they didn't pay me to say this. I still have my criticisms of Grok, but holy shit, this is what UX is supposed to be like. xAI fuckin cooked.