I can say that the proprietary LLMs got really bad fast. The higher number version should mean better, but I'm getting worse results more often. I hope this is due to AI poison, AI data feeding, and the models limitations.