Detailed aspects versus GPT-4 turbo 1. World knowledge, slightly inferior. GPT-4 turbo was able to detail a protagonist's childhood year by year for a Japanese novel with near 100% accuracy (That a human reader would get the chronology wrong). Gemini ultra much more easily confused.
2. Creativity, Gemini ultra wins. Its writing style has far more flair than GPT-4 turbo, it also occasionally made some stunning analysis that I never thought of and made perfect sense. GPT-4 turbo is more like a textbook repeater, it doesn't make many mistakes, but also rarely surprises you with anything original.
3. Accuracy, GPT-4 turbo still makes fewer mistakes. Including in subtle logic (Like having a hypothetical battle between two characters in the same universe, considering the strengths and weaknesses of their powers, etc).
So this is definitely Google's first real-deal LLM. Its not better than current GPT-4 turbo, but its getting there. OpenAI must be feeling the fire to release GPT-5 before the end of the year.