That's not the direction of LLM progress these past 2 years. Cost has always decreased per unit intelligence. Open weight models with lower parameters beat gpt4 that we were so impressed by.