I know this rattles a lot of people's jimmies but in my own personal benchmarks (three problems a 10 year old could easily solve), all "frontier" LLMs are still getting the same output quality as GPT4 or even GPT3. So at least for me cheapest provider wins. No, I'm not sharing the problems with you. Unlike a certain influencer who really likes pelicans, I don't get paid for boosting these companies.