What “better model”? If you’re referring to whatever LLM GPT-4 was before RLHF, by what measure might this model be better? As I pointed out above, what OpenAI is selling is “best” for their customers.